All updates
Every new model, feature, capability, setting and plugin recorded since 1 September 2026. Filter by AI, type, model and date.
November 2026
Grok Bot phone calls are limited to eligible US accounts and US numbers
An official post dated November 5 confirms that Grok Bot can make real phone calls, handle menus, hold times, and transfers, and provide live transcripts. Users can respond when needed and receive a summary at the end of the call; the rollout is still ongoing and is currently limited to eligible US accounts calling US numbers.
Usage details
Who can use it: Eligible US accounts; US numbers only; gradual rollout. The post does not confirm access for accounts or numbers outside the US.
How to use it: Update the Grok Bot app; tell Bot whom to call and what you want it to do; respond in the chat box if it asks for information during the call.
Tip: Before the call, clarify the name, account details, and desired outcome for Bot; handle final approvals for charges, cancellations, or contracts yourself.
✓ Verified by official sourceSource: Grok’un 5 Kasım 2026 tarihli resmi gönderisiLocal Docker recipe added for /verify-pr
Qwen Code published a local `/verify-pr` Docker recipe that runs untrusted PR code in a container. Model credentials stay on the host, behind a chat-completion proxy.
Usage details
Who can use it: Developers using a local Qwen Code installation, especially those who want to isolate untrusted PRs.
How to use it: Use the local `/verify-pr` Docker recipe in the Qwen Code repository; keep PR code in the container and model credentials on the host proxy.
Tip: Limit the container running PR verification to only the necessary file and network permissions.
✓ Verified by official sourceSource: Qwen Code v0.26.7 sürüm notlarıLinked PR events arrive in the session as untrusted prompts
GitHub events for linked PRs are now passed to the relevant session as untrusted prompts. Responses remain private until explicitly published.
Usage details
Who can use it: Users who link GitHub PRs to Qwen Code Channel sessions.
How to use it: Link a PR to a Channel session; review incoming GitHub events as untrusted prompts and publish the response only when needed.
Tip: Treat automatically received PR content as external data to verify, not as instructions.
✓ Verified by official sourceSource: Qwen Code v0.26.7 sürüm notlarıCatalog API /batch endpoint removed in v27
The Catalog Batch API endpoint is deprecated in v27.0+ and no longer works. Integrations need to migrate to the /items_batch endpoint; /batch will continue to work in v26.0 and earlier until support ends.
Usage details
Who can use it: Catalog API integrations, advertiser businesses, and agencies using Marketing API v27.0+.
How to use it: Replace POST /{catalog-id}/batch calls with POST /{catalog-id}/items_batch and test according to the migration guide.
Tip: Before moving your API version to v27, test and compare the requests and error responses of both endpoints.
✓ Verified by official sourceSource: Graph API ve Marketing API v27.0 ChangelogSource: Graph API v27.0 ve Marketing API v27.0 duyurusuFilterable live timeline added for Channel events
A separate page showing Channel events has been added to Web Shell. It presents events, skip reasons, replies, task transitions, and polling results in a filterable, automatically refreshed timeline.
Usage details
Who can use it: Teams using Qwen Code Web Shell Channel.
How to use it: Open the Channel events page in Web Shell; use the filters and follow the timeline with live refresh enabled.
Tip: When investigating the source of a channel issue, narrow the skip reason and task transition filters first.
✓ Verified by official sourceSource: Qwen Code v0.26.7 sürüm notlarıChatGPT Images 2.6 brings faster generation and more consistent editing
OpenAI made ChatGPT Images 2.6 available on November 4. The model generates images faster than the previous version while better preserving elements such as characters’ clothing and product details during edits; OpenAI says generation is 50% faster.
Usage details
Who can use it: Rolling out worldwide across all subscription tiers in ChatGPT and Codex on web, desktop, and mobile.
How to use it: In ChatGPT, ask for an image in a conversation or open Images via More > Images; upload an existing image and describe the edits you want. You can also use the image generation and editing workflow in Codex.
Tip: Explicitly lock in character, product, and brand details in your editing prompt; asking for one major change per turn can improve consistency.
✓ Verified by official sourceModels: Images 2.6Source: ChatGPT sürüm notlarıSource: Images 2.6 tanıtımıSource: Images in ChatGPT yardım sayfasıNew sign-ups and downloads for Conductor have stopped; Cloud shuts down on November 20
According to Conductor’s official announcement, new sign-ups and app downloads stopped on November 4 following its move to Meta; existing free users cannot upgrade to Pro, while existing Pro users can use the features for free. Conductor Cloud will become read-only on November 13, shut down on November 20, and begin deleting cloud data. Since the source is a Conductor announcement with TechCrunch confirmation, rather than an official Meta channel, verified=false.
Usage details
Who can use it: Conductor users; the local Mac app will continue to work, while Cloud users’ migration timeline depends on the November 13–20, 2026 schedule.
How to use it: Right-click the Cloud workspace and select Transfer to local; use Settings → Account → Export data to export all data, and do so before November 20.
Tip: Move your cloud workspace and chat archive to a local copy right away; waiting until the read-only phase may limit your export options.
DeepSeek Harness v0.2.2-rc.1 released
The first release candidate in the DeepSeek Harness 0.2.2 series has been released. It covers user and developer changes since v0.2.1-rc.5, including plugins, mobile use, automation, SDK, and desktop workflows.
Usage details
Who can use it: Available in the Web interface and desktop app as part of Public preview; desktop distribution is available for macOS and Windows.
How to use it: To update from source, pull the git repository and run the npm/pnpm installation, or launch it with the npx @deepseek-ai/dsh web command in the README; desktop users can use the installer on the official download page.
Tip: Because the release candidate may include breaking changes, test it in a separate profile and workspace before using it in production workflows.
deniedMcpServers URL blocking also covers connectors
URL entries also block matching Anthropic servers, cloud connectors, and runner connectors.
Usage details
Who can use it: Organizations managing MCP and connector policies.
How to use it: Add a URL or connector name to deniedMcpServers in managed settings.
Tip: Verify broad URL matches in a test group first.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesDesktop subagents can control the built-in browser with permission
Running subagents can now use Desktop’s built-in browser with user permission. This lets parallel tasks handle page navigation and verification separately from the main Agent flow.
Usage details
Who can use it: Kimi Code Desktop v1.3.0; subagent tasks that require user approval and browser access.
How to use it: Update Desktop to v1.3.0, assign the subagent a task that includes browser verification, and approve access in the permission card that appears.
Tip: Use Always Ask for initial trials, and check which pages the subagent accesses before the task is complete.
✓ Verified by official sourceSource: Kimi Code Desktop 1.3.0 resmi changelog’uSource: Kimi Code Desktop görev yürütme kılavuzuMinimum supported Node.js version raised to 24.9
The minimum supported Node.js version for DeepSeek Harness has been raised to 24.9; update Node.js before upgrading if you’re using an older runtime.
Usage details
Who can use it: DeepSeek Harness installations run from source code or via CLI; Node.js 24.9 and later.
How to use it: Check the version with node --version. If it’s earlier than 24.9, upgrade Node.js, then reinstall DeepSeek Harness dependencies.
Tip: Before switching to the candidate version, also verify that Node.js 24.9 is available in CI and remote SSH environments.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıDots rolling out for general availability in Enterprise, Edu, and Healthcare workspaces
Dots is becoming generally available in eligible ChatGPT Enterprise, Edu, and Healthcare workspaces; the administrator preview continues through November 18, 2026, and access is off by default during this period. Dots supports data residency and Enterprise Key Management with EU/US inference residency, and may fall under HIPAA in eligible workspaces; workspaces with UAE data or inference residency are not supported.
Usage details
Who can use it: Eligible Enterprise, Edu, and Healthcare workspaces; administrator preview from 2026-11-04–2026-11-18, followed by default enablement unless the administrator opts out. Excludes workspaces with UAE data or inference residency.
How to use it: As an administrator, open Workspace settings > Permissions & roles > Permissions > Workspace default > Workspace capabilities, review Use dots and the relevant computer capabilities, and save. If needed, keep on-request approval and auto-review enabled under Admin Console > Agent Security > Global > Requirements. Members can then create their own dot through the onboarding flow in the mobile app, desktop app, or desktop web.
Tip: Since access is off during the preview, first enable it only for the roles and members who need it; regularly review personal dots’ connections, usage, and administrator activity.
✓ Verified by official sourceModels: GPT-6 AstraSource: ChatGPT Enterprise ve Edu sürüm notlarıSource: ChatGPT çalışma alanlarında dots yönetimiSource: Dot ile çalışmaya başlamaInvalid claude.json is no longer overwritten with defaults
An invalid ~/.claude.json is no longer replaced with defaults; startup fails and the file is preserved.
Usage details
Who can use it: CLI installations using ~/.claude.json.
How to use it: Restore from the recommended backup and restart.
Tip: Make a copy of claude.json before startup.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesGPT-Image-2.6, complexity, SVG, GIF, and markup editing come to the Images API
OpenAI released the GPT-Image-2.6 Sunburst and Flare models in its image generation and editing API. The Image API now offers complexity control, SVG and animated GIF output; markup-guided editing has also been added to the supported GPT Image 2.5 and 2.6 models.
Usage details
Who can use it: For organizations using the API, on the `v1/images/generations` and `v1/images/edits` endpoints; rate limits vary by usage tier, and organization verification may be required.
How to use it: Set the `model` field to `gpt-image-2.6-sunburst` or `gpt-image-2.6-flare` in a `POST /v1/images/generations` or `/v1/images/edits` request. Choose `low`, `medium`, or `high` for the `complexity` field; use `output_format: "svg"` or `output_format: "gif"` when you need SVG or GIF output. For GIF generation, you can also set the `frame_count` and `frame_duration_ms` fields.
Tip: Try `complexity: "low"` for focused edits and `medium` or `high` for scenes with dense text or many objects; higher complexity may increase latency.
✓ Verified by official sourceModels: gpt-image-2.6-sunburst, gpt-image-2.6-flareSource: OpenAI API değişiklik günlüğüSource: GPT-Image-2.6 Sunburst model sayfasıSource: GPT-Image-2.6 Flare model sayfasıChange the model and thinking level while Kimi Code is running
The `/model` and `/effort` commands can now be used while the Agent is running. The model call already in progress keeps its existing settings; the selected value applies to the next turn, and Compact and continue waits for the current turn to finish.
Usage details
Who can use it: Kimi Code CLI v2.5.0 and later.
How to use it: During a running turn, enter `/model` or `/effort` and select a new model or the `low`, `high`, or `max` level.
Tip: When changing the model midway through a long turn, keep in mind that it affects the next turn, not the current result.
✓ Verified by official sourceSource: Kimi Code CLI 2.5.0 resmi changelog’uSource: Slash Commands referansıKimi Code CLI runs remote environments over SSH, Docker, and custom commands
Agent can now run file and Shell operations on an SSH machine, in a Docker-compatible container, or through a custom launcher command; the CLI, model requests, credentials, and session state remain local. Windows remote targets are also supported over SSH and custom commands.
Usage details
Who can use it: Kimi Code CLI v2.5.0; SSH, Docker, or custom-command environments; Docker is not supported on Windows targets.
How to use it: Define `[environments.dev-box]` in `~/.kimi-code/config.toml`, then run `kimi --environment dev-box` or `kimi --environment dev-box --environment-cwd /path` for a different directory.
Tip: Install the Kimi Code executor on the remote machine in advance; commands don’t silently fall back to the local machine if the connection drops.
✓ Verified by official sourceSource: Kimi Code CLI 2.5.0 resmi changelog’uSource: Remote environments kılavuzuKimi Code Desktop lets you add multiple working directories to projects
With Desktop v1.3.0, you can select a primary directory and additional source folders when creating a project. The directory list can be edited later, and older additional directory configurations can be imported.
Usage details
Who can use it: Kimi Code Desktop v1.3.0; local multi-directory projects.
How to use it: When creating a new project, select a primary directory and add source folders; later, edit the project directory list or import a legacy additional directory configuration.
Tip: Choose the most stable root as the primary directory for shared configuration and conflict resolution.
✓ Verified by official sourceSource: Kimi Code Desktop 1.3.0 resmi changelog’uSource: Kimi Code güncel değişiklik özetiKimi Code’s plugin marketplace and built-in MCP OAuth flow updated
Kimi Code 2.5.0 supports plugin marketplace interactions and built-in OAuth authorization for MCP servers within plugins. Official, Curated, and custom URL sources can all be used in the same manager.
Usage details
Who can use it: Kimi Code CLI v2.5.0 and Desktop plugin settings; plugins are user-scoped.
How to use it: Open `/plugins` in the TUI and install from the Official or Curated tab; use `/plugins marketplace <source>` for a custom source, and start the `/mcp-config` flow for MCP OAuth.
Tip: After installing or enabling a plugin, use `/reload` or `/new` to apply the change to a new session.
Kimi Code can fork from a previous turn or a running session
A session can now be copied independently from a selected previous turn; a fork can also be created while the source session is running. This makes it possible to try alternative implementation approaches without disrupting the current flow.
Usage details
Who can use it: Kimi Code CLI v2.5.0 and Kimi Code Desktop v1.3.0.
How to use it: Use `/fork` in the CLI; in Desktop, select `Fork from here` from the session or previous-turn menu.
Tip: Make major experimental changes in a fork to keep the main session’s context and goal clear.
Kimi Code REST task and status endpoints removed
`/sessions/{id}/tasks`, status and snapshot endpoints, along with the legacy WebSocket subscribe/unsubscribe flow, have been removed. Task output and control have moved to WebSocket `task.watch`, `task.unwatch`, `task.stop` and `task.background` requests.
Usage details
Who can use it: Third-party clients using Kimi Code’s local REST/WebSocket API; after 2.5.0.
How to use it: Remove legacy REST task calls and monitor tasks with `task.watch` and `task.unwatch` over `/api/v1/ws`; send `task.stop` and `task.background` to stop a task or move it to the background.
Tip: In migration code, check the server version and capabilities via `/api/v1/meta`, and show users a clear error message for legacy endpoints.
✓ Verified by official sourceSource: Kimi Code CLI 2.5.0 resmi changelog’uSource: Kimi Code Server API referansıKimi Code WebSocket stream carries Agent and task status
The local server WebSocket now sends live status, settings, context usage and task changes through `agent`, `usage`, `goal`, `todo`, `task.list` and `task` frames. The subscription flow has been updated to use `subscribe_v2` and resubscribe after reconnecting.
Usage details
Who can use it: Integration developers using the Kimi Code CLI 2.5.0 local REST/WebSocket server.
How to use it: Start `kimi web`; read the live `/openapi.json` and `/asyncapi.json` schemas for your version and use `subscribe_v2` over `/api/v1/ws`.
Tip: Don’t bind your client to a fixed list of fields; reread the live OpenAPI and AsyncAPI schemas with every Kimi Code version.
✓ Verified by official sourceSource: Kimi Code CLI 2.5.0 resmi changelog’uSource: Kimi Code Server API referansıKimi K3.1 Pro open weights released on Hugging Face
Moonshot AI has released Kimi K3.1 Pro’s full model weights, making it possible to run the model on your own infrastructure. The official model card says it uses 2.8 trillion total parameters, 104 billion active parameters, and a 1,048,576-token context window; it recommends vLLM, SGLang, and TokenSpeed.
Usage details
Who can use it: Available to developers and researchers via Hugging Face; the Kimi K3.1 Pro License terms apply.
How to use it: Download the `moonshotai/Kimi-K3.1-Pro` repository from the Hugging Face model card and serve it using one of the vLLM, SGLang, or TokenSpeed examples in the card.
Tip: Before putting a 1-million-token context window into production, use a short pilot to measure GPU, latency, and memory costs on your own setup.
✓ Verified by official sourceModels: Kimi K3.1 ProSource: Kimi.ai’nin ağırlık yayın duyurusuSource: Kimi K3.1 Pro resmi Hugging Face model kartıcountry and location_spec rejected when creating Lookalike audiences
In Marketing API v27.0+ custom audience creation requests, lookalike_spec.country and lookalike_spec.location_spec are no longer accepted. Lookalike audiences get their delivery locations from the ad sets targeting them; existing audiences and campaigns are unchanged.
Usage details
Who can use it: Marketing API v27.0+; creating new Lookalike audiences with POST /act_{ad-account-id}/customaudiences.
How to use it: Remove the country and location_spec fields from lookalike_spec in new custom audience requests; define the location in ad set targeting.
Tip: Validate your new audience creation code against the v27 endpoint first, without changing existing audiences.
✓ Verified by official sourceSource: Graph API ve Marketing API v27.0 ChangelogSource: Graph API v27.0 ve Marketing API v27.0 duyurusuManaged Agent supports full model configuration at session startup
Managed Agent now accepts the full model configuration when a session is created. This allows you to specify not only the model name but also other model options the session needs in the initial request.
Usage details
Who can use it: Developers using the Qwen Code Managed Agent session creation flow.
How to use it: Send the full model configuration in a new Managed Agent session creation request; populate the model fields in the session creation schema for the version you’re using.
Tip: Setting model options at session startup helps ensure consistent behavior in reconnection and supervised generation flows.
✓ Verified by official sourceSource: Qwen Code v0.26.7 sürüm notlarımanaged-agents-onboard applies risk judgment by default
managed-agents-onboard now grants the auto permission policy to new agents and quickstart MCP tools by default; you can opt out with run without asking.
Usage details
Who can use it: Managed Agents onboarding and Claude Console workflows.
How to use it: Run /claude-api managed-agents-onboard; keep the secure default or choose run without asking.
Tip: Use run without asking only for low-risk tasks.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesAWS assume_role support for Mantle upstreams
Mantle upstreams support assuming a role in another AWS account and developer-based AWS cost attribution.
Usage details
Who can use it: Claude apps gateway administrators and AWS setups using a Mantle upstream.
How to use it: Add the assume_role field and target role to the Mantle upstream configuration.
Tip: First test the role’s trust relationship and cost permissions with a limited scope.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesmeta_moment_maker_spec being retired in third-party apps
In Marketing API v27.0+, meta_moment_maker_spec is no longer returned in GET responses for third-party apps; adding the field to create or update requests results in a version error. Use product_type for Ads Burst and Ads Super Burst inventory; saturation mode is set by the server.
Usage details
Who can use it: Marketing API v27.0+; ad set and reach frequency prediction integrations.
How to use it: Remove meta_moment_maker_spec from read and write flows; use product_type=ADS_BURST or ADS_SUPER_BURST in reach/frequency prediction requests.
Tip: Validate the new product_type flow in a separate test app before the older API version expires.
✓ Verified by official sourceSource: Graph API ve Marketing API v27.0 ChangelogApproved shell commands run in the correct folder after restart
After a restart, the approved command now runs in the requested folder rather than the starting folder; it won't run if the folder doesn't exist.
Usage details
Who can use it: Cloud and self-hosted runner sessions.
How to use it: Approve the command again after restart and check the working folder.
Tip: Verify the folder before running destructive commands.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesThe Page Insights total_video_consumption_rate metric is no longer returned
According to Meta’s November 4 v27 announcement, the total_video_consumption_rate metric has not been returned by any Graph API version since early November 2026. No direct replacement for the metric was provided.
Usage details
Who can use it: All Graph API versions; GET /{video-id}/video_insights.
How to use it: Remove the total_video_consumption_rate field from video_insights requests and handle its absence without treating it as an error.
Tip: Instead of treating an empty or missing metric as zero, show an ‘unavailable’ status in reports.
✓ Verified by official sourceSource: Graph API ve Marketing API v27.0 ChangelogSource: Graph API v27.0 ve Marketing API v27.0 duyurusuProject hooks require explicit folder or workspace trust
Qwen Code now requires explicit folder or workspace trust before running repository-sourced hooks, permission grants, and project language servers. If trust isn’t granted, a warning appears in the interactive session.
Usage details
Who can use it: All interactive Qwen Code sessions that use `.qwen/settings.json` or project hooks in a repository.
How to use it: Review the trust prompt when you open a project; grant folder/workspace trust only to a folder whose source you’ve verified, then run the hooks.
Tip: Before granting trust in unknown repositories, inspect `.qwen/settings.json` and the language server configuration.
✓ Verified by official sourceSource: Qwen Code v0.26.7 sürüm notlarıPTC execution architecture and SSH protocol updated
Native and PTC tool execution have been split into Agent-scoped plugins; PTC is enabled through a configuration overlay. The PTC Node communication channel has moved to birpc, the SSH protocol has been upgraded from 2 to 3, and run_code returns ABORTED when canceled by the caller.
Usage details
Who can use it: Developers who distribute their own bootstrap or SSH helper, and SDK/Session consumers using PTC.
How to use it: Update the self-hosted bootstrap and SSH helper components together, enable PTC through a configuration overlay, and add ABORTED to the SDK/Session cancellation class.
Tip: Upgrade the protocol on the client and remote helper in the same deployment; do not mix an old helper with a new client.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıRemote connection settings for qwen serve without a restart
Qwen Code added the `Settings → Connections → Allow remote connections` setting. It lets you manage remote access, token usage, and client disconnections without restarting the service.
Usage details
Who can use it: Qwen Code installations using `qwen serve`.
How to use it: Enable `Settings → Connections → Allow remote connections`; configure a token for remote client access and manage the connection without restarting.
Tip: Enable remote access only for trusted networks and clients; do not put the token in source code or a shared configuration file.
✓ Verified by official sourceSource: Qwen Code v0.26.7 sürüm notlarıRemote Control and runner can start new sessions
Fixed an issue that left Remote Control, the daemon, and the runner with spawn ENOENT when an update or version cleanup removed the binary.
Usage details
Who can use it: Remote Control, daemon, and self-hosted runner installations.
How to use it: Upgrade the host, daemon, and runner components to 2.1.318.
Tip: Plan version cleanup around the lifecycle of running binaries.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesRunner orchestrator log fields have changed
ok lines now include a reason field at every stdout level, and nack lines use a new reason position.
Usage details
Who can use it: Operations teams that parse runner logs.
How to use it: Update log parsers to match the reason field and stdout format.
Tip: Replace old retryable patterns with ones matching the new format.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesSession API provides snapshot and continuous event subscriptions
Session API provides separate interfaces for reading snapshots and continuously subscribing to session event logs, execution status, and event projection data.
Usage details
Who can use it: SDK and plugin developers using DeepSeek Harness Session API.
How to use it: Call the snapshot read endpoint in Session API; then subscribe to the event log and execution status, and associate incoming events with the Session ID.
Tip: Add reconnection and resume-from-last-event logic to continuous subscriptions to prevent event loss during brief network outages.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıTencent Hy4 announced: visual understanding and computer use for the first time
Tencent announced its next-generation large language model, Hy4. For the first time, the model adds visual understanding (fine-grained image processing, complex chart analysis, mathematical visual reasoning) and computer use through the command line and graphical interfaces; it also improves instruction following, long context, and long-running agent tasks. It ranks 14th in the unique rankings with a score of 35 on the Artificial Analysis Intelligence Index v5.0.1 (1M-token context, text+image input).
Usage details
Who can use it: For now, it is available through Tencent WorkBuddy and WeCom integration; according to Artificial Analysis, there is no public API. Tencent says it will release the weights and system cards as open source “in the coming days.”
How to use it: WorkBuddy users can use Hy4 directly for office, visual design, 3D scene, and web development tasks; when connected to WeCom, it tracks projects using chat, email, document, calendar, and meeting data.
Tip: When the weights are published on Hugging Face (huggingface.co/tencent), check the license terms before trying them on your own infrastructure; for now, treat it as a closed model.
✓ Verified by official sourceModels: Tencent Hy4Source: Tencent Launches Hy4 (PR Newswire)Source: Tencent Hy4 – Artificial AnalysisThe trusted_proxies guidance is now safer
Log hints say that only proxies that add X-Forwarded-For should be listed in trusted_proxies.
Usage details
Who can use it: Administrators running a gateway behind a proxy.
How to use it: Limit listen.trusted_proxies to trusted proxies.
Tip: Don’t trust the client header directly.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesTypeScript and Python SDKs expanded with Web capabilities
The SDKs support file delivery, questions and callbacks, scheduling, skills, and context capabilities to better match Web. Plan review, goal persistence, and action approvals have also been added; plugin changes take effect when the SDK restarts.
Usage details
Who can use it: Developers using the DeepSeek Harness TypeScript and Python SDKs.
How to use it: Install the latest SDK, enable file delivery or question callbacks, and use the plan/action approval flow; restart the SDK process after changing a plugin.
Tip: To compare Web and SDK behavior, run the same session plan in both clients and test callback and approval records.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıUCP-powered direct checkout expands to Canada in the Gemini app
Direct checkout powered by Universal Commerce Protocol (UCP) is beginning to roll out to users in Canada in Google Search AI Mode and the Gemini app. For eligible listings, users can buy directly from select Canadian retailers using Google Pay.
Usage details
Who can use it: Canada; Gemini app and Search AI Mode; initially, eligible retailers such as lululemon, select Shopify merchants, Shoppers Drug Mart, and Simons.
How to use it: Search for an eligible product or retailer in the Gemini app or Search AI Mode; select the direct checkout flow shown in the results and complete the purchase using payment and delivery details saved in Google Pay.
Tip: Verify your payment and shipping details in Google Wallet in advance; eligibility varies by retailer and product.
✓ Verified by official sourceSource: Google Canada — Kanada’da agentic shopping ve UCP destekli checkoutDevice expansion enabled by default for eligible Sales ad sets
Starting with v27.0, for eligible new Sales ad sets targeting a specific device or operating system, Reach more people across other mobile devices is enabled by default if the device field is omitted, and is saved as device=1. Existing ad sets are unchanged; Mobile App conversion location and Special Ad Categories are excluded.
Usage details
Who can use it: Marketing API v27.0+; eligible new Sales ad sets targeting a device or OS that don’t use Mobile App conversion location.
How to use it: To deliver only on the devices and operating systems you selected, send targeting.targeting_automation.individual_setting.device=0.
Tip: To prevent delivery from silently expanding, explicitly set the device value to 0 or 1 in new ad sets.
✓ Verified by official sourceSource: Graph API ve Marketing API v27.0 ChangelogSource: Graph API v27.0 ve Marketing API v27.0 duyurusuWeb Fetch no longer rejects proxy and internal network addresses
Web Fetch can also read non-public addresses and allows proxy or internal network use. Since restricting network access is left to the deployment, security policies must be configured separately.
Usage details
Who can use it: Self-hosted deployments or deployments behind a proxy that use Web Fetch; internal network addresses may be accessible.
How to use it: Configure the deployment’s network egress policy before using Web Fetch, then test an allowed internal network or proxy address.
Tip: Don’t leave internal network access enabled by default; implement allowlist, DNS rebinding, and SSRF checks.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıSix legacy features blocked in new Graph API apps
Starting November 4, new apps can’t use multi-ID requests, date/time formatting, HTTP method override, field aliases, JSONP callbacks, or asynchronous batch requests; existing apps also can’t add new use cases for them. Existing use in older versions is supported for now; the remaining features will be removed in Graph API v28.0.
Usage details
Who can use it: All Graph API versions; new apps and new use cases in existing apps.
How to use it: Scan your code for uses of ids, date_format, method override, .as(...), callback, and asyncbatch; migrate to supported standard request flows before upgrading to v28.
Tip: Search your codebase for these six parameters before App Review or a version upgrade, and prevent new usage from being added.
✓ Verified by official sourceSource: Graph API ve Marketing API v27.0 ChangelogSource: Graph API v27.0 ve Marketing API v27.0 duyurusu--continue preserves conversations after a broken tool call
Fixed an issue where the --continue command lost the previous conversation after retries for a broken tool call were exhausted.
Usage details
Who can use it: Sessions using --continue in the Claude Code CLI.
How to use it: Update to 2.1.318 and run claude --continue.
Tip: Keep transcript backups for long-running tasks, too.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesPolitical and social issue ads can run again in the US
According to two ad industry sources citing Meta’s official Government and Nonprofits announcement, new US ads about social issues, elections, and politics, as well as edits to existing ads, were re-enabled at 00:00 PT on November 4, 2026. The change is marked as unverified because Meta’s official page could not be accessed as text in this review.
Usage details
Who can use it: Accounts advertising about social issues, elections, or politics in the US; after 00:00 PT on November 4, 2026.
How to use it: Create a new ad or edit an existing one in Ads Manager after that time; separately check authorization, the Paid for by disclaimer, and current Advertising Standards requirements.
Tip: Reopening does not mean all ads will be automatically approved; verify review and disclaimer status before publishing.
Failed settings changes in network home directories are no longer reported as successful
Logout, MCP, and permission changes that can’t be written to a shared network home directory no longer appear to succeed silently.
Usage details
Who can use it: CLI installations using a shared network home directory.
How to use it: Upgrade to 2.1.318 and rerun the command; fix the write issue.
Tip: Check that configuration files are writable.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesConfiguration saves in network home directories are more reliable
Reduced the likelihood of claude.json, settings, and saved logins being deleted, emptied, or replaced with an older copy.
Usage details
Who can use it: Installations using a shared network file system.
How to use it: Upgrade the client to 2.1.318, then save your settings and login.
Tip: Back up critical configuration separately.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesSettings that appear missing on network file systems are checked again
If a network file system mistakenly reports that the configuration is missing, the session checks again; this reduces the loss of MCP and trusted folders.
Usage details
Who can use it: Sessions using a network file system.
How to use it: Update to 2.1.318 and restart the session.
Tip: Monitor file visibility delays.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesAirOps added to Claude Marketplace
AirOps offers an agent-based platform for AI search strategy, content, and multichannel execution. It helps marketing teams connect brand performance in AI search with revenue and pipeline outcomes.
Usage details
Who can use it: Partner agent/product listed on Claude Marketplace; enterprise purchasing and an AirOps account are required.
How to use it: Open the AirOps listing on Claude Marketplace and submit a purchase request; connect your brand, content, and multichannel campaign goals during partner setup.
Tip: For your first pilot, pair AI search visibility with a measurable pipeline metric.
✓ Verified by official sourceSource: Claude Marketplace’te güvenlik, üretkenlik ve yaratıcı çalışma için yeni ajanlar ve ürünlerIncorrect startup block caused by allowedProviders fixed
Fixed an issue where an allowedProviders list saved by another login incorrectly rejected session startup.
Usage details
Who can use it: Sessions using multiple providers on one machine.
How to use it: Upgrade to 2.1.318 and restart the session.
Tip: Keep provider permissions aligned with your organization’s policy.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesUnified gRPC and HTTP streaming arrives for AlphaGenome
AlphaGenome (alphagenome-004) now supports unified HTTP/2 gRPC streaming and chunked binary protobuf HTTP streaming on Gemini Enterprise Agent Platform. The update enables lower-latency 1 MB track predictions across seven DnaModelService methods, concurrent multi-variant evaluation, and model fold selection; existing `alphagenome-003` HTTP deployments remain supported.
Usage details
Who can use it: Gemini Enterprise Agent Platform; AlphaGenome access requires an allowlist, and self-deployed endpoints require suitable Google Cloud GPU capacity.
How to use it: On your AlphaGenome endpoint, use `PredictionService.StreamDirectRawPredict` for gRPC (shared endpoint, port 8081) or `streamRawPredict` for HTTP (dedicated endpoint, port 8080); use `alphagenome_cloud_sdk` to configure `ALL_FOLDS` or select a specific fold.
Tip: For large variant scans, use concurrent `score_ism_variants` calls and gRPC binary streaming to reduce the risk of HTTP gateway timeouts.
✓ Verified by official sourceModels: AlphaGenome (alphagenome-004), alphagenome-003Source: Google Cloud sürüm notları — 4 Kasım 2026Source: AlphaGenome — Gemini Enterprise Agent Platform dokümantasyonuAnthropic-compatible requests preserve configured timeouts
Anthropic-compatible requests now honor configured transport timeout and dispatcher values. This prevents requests from being cut off early when a longer API timeout is set.
Usage details
Who can use it: Developers using Anthropic-compatible provider endpoints in Qwen Code.
How to use it: Set the desired transport timeout in the Anthropic-compatible provider configuration; verify that the value is preserved with a long-running request.
Tip: When increasing the timeout, also monitor the provider’s actual server limit and the total task budget separately.
✓ Verified by official sourceSource: Qwen Code v0.26.7 sürüm notlarıA session without a repository in an Anthropic-hosted cloud environment
Cloud sessions can now be started without selecting a repository by choosing No repository.
Usage details
Who can use it: Anthropic-hosted cloud environments.
How to use it: Select No repository when choosing a repository during session creation.
Tip: Use this option for tasks that don’t require a repository.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesTools can be enabled based on step conditions
Tools can now be enabled conditionally at the step level; model notification, PTC context, and execution permission follow the same decision. Plugin developers need to migrate the tools.resolve, tools.executable, and prompt-merging integrations.
Usage details
Who can use it: DeepSeek Harness plugins and Agent tools; a new tool resolution contract for developers.
How to use it: In the plugin, tie tool resolution to the step condition, update the tools.resolve and tools.executable integrations, and test model notification and execution permission together.
Tip: For conditional tools, restrict the tool schema shown to the model using the same condition as execution.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıBackground subagents move shell jobs to the background after three minutes
Background subagents move shell commands that run for three minutes to the background across all providers.
Usage details
Who can use it: Background subagents and all supported providers.
How to use it: Assign a long-running shell job to a subagent and monitor its background status.
Tip: Redirect long command output to a file.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesBackground tasks show live notifications and detailed output
The CLI displays a notification while a background task is producing a result; Desktop brings running tasks to the foreground and accumulates detailed output live. Agent and subagent status, model, and result information can be tracked in the task view.
Usage details
Who can use it: Kimi Code CLI 2.5.0 and Desktop v1.3.0.
How to use it: Open `/tasks` in the CLI; in Desktop, select the running task from the background work bar above Composer or the Agent/Subagent panel.
Tip: After moving a long task to the background, check its live status and output first rather than starting the same task again.
Asana AI Teammates added to Claude Marketplace
Asana AI Teammates offers ready-to-use agents that use Asana Work Graph context and have separate identities, comprehensive permissions, and audit trails. They can route decision points in multistep work to the responsible person.
Usage details
Who can use it: Partner agent listed on Claude Marketplace; an Asana workspace, enterprise purchasing, and permission configuration are required.
How to use it: Open the Asana AI Teammates listing on Claude Marketplace and submit a purchase request; configure the agent identity and the scope of Asana it can access with limited permissions.
Tip: Keep the audit trail readable by defining a single workflow and a narrow permission scope for each teammate.
✓ Verified by official sourceSource: Claude Marketplace’te güvenlik, üretkenlik ve yaratıcı çalışma için yeni ajanlar ve ürünlerBehavior for continuing after skipped questions and partial skill discovery fixed
It is now clear that when a user skips a question, the model should continue without repeating that question, and that skipping does not count as permission. If a skill root cannot be read, skills in healthy roots continue to be used, and incomplete discovery is reported.
Usage details
Who can use it: DeepSeek Harness question flows and profiles using multiple skill roots.
How to use it: Choose skip on a question card and observe the flow continue; temporarily make one skill root inaccessible to verify that the remaining skills are listed.
Tip: Treat a skipped question as missing a decision rather than as permission, and use a separate approval policy for sensitive actions.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıCaching and background notifications improved in Auto review
Auto review makes better use of cached historical context between consecutive tool calls. When a background job completes, small outputs are included directly in the notification, and duplicate notifications for previously read results are reduced.
Usage details
Who can use it: DeepSeek Harness Auto review and background tasks.
How to use it: Run consecutive tool calls with Auto review; when a task completes, open the small output in the notification and check that previously read results are not repeated.
Tip: For repeated tool calls, preserve the same context and avoid starting unnecessary new sessions to increase cache gains.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıAuto permission and approval policy is preserved on shutdown
When the app closes, the Auto session's permission and approval policy no longer accidentally switches to Full access; the saved policy is preserved.
Usage details
Who can use it: Auto sessions and saved permission/approval policies.
How to use it: Set the permission policy in an Auto session, close and reopen the app, then verify the previous policy value in the session settings.
Tip: Check that the permission policy has the expected value before restarting automated tasks.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıAuto-update immediately retries version checks after network errors
A brief network interruption during version validation is now retried without waiting for the next check.
Usage details
Who can use it: Claude Code installations with Auto-update enabled.
How to use it: Leave Auto-update enabled and monitor checks after brief network interruptions.
Tip: Allow update domains through your corporate proxy.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesSigning in again from the same home directory no longer interrupts jobs
Fixed an issue where running jobs ended with Not logged in when another terminal signed in again with the same account.
Usage details
Who can use it: Multiple terminals or hosts sharing the same home directory.
How to use it: Use 2.1.318 on all hosts and resume the job.
Tip: Keep versions consistent on shared machines.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesFailed Remote Control sessions no longer appear available
Fixed an issue that made a /remote-control session appear available after the connection failed.
Usage details
Who can use it: CLI sessions using Remote Control.
How to use it: Reopen the /remote-control menu and check the status.
Tip: Start a new connection instead of using the failed card.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesBedrock CountTokens error identifies the right component
The warning says Claude Code handles counting after a 501, and IAM should be checked only for access-denied errors.
Usage details
Who can use it: Gateways using a Bedrock upstream.
How to use it: For a CountTokens error, read the warning and check IAM only for access-denied errors.
Tip: Distinguish 501 errors from permission denials.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesEmpty claude.json is treated like a missing file
An empty ~/.claude.json without a backup is no longer reported as corrupt, and startup proceeds as if the file were missing.
Usage details
Who can use it: Installations with an empty configuration file.
How to use it: Leave the file empty and start with 2.1.318.
Tip: Verify that organization settings are reloaded.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesIdle compaction now replaces the conversation with a summary
A bug where idle compaction replaced the conversation with a non-summary response has been fixed.
Usage details
Who can use it: VS Code, Desktop, and CLI sessions with idle compaction enabled.
How to use it: Leave a long session idle and check the generated summary.
Tip: Save important decisions separately in a short file.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesA malformed NODE_EXTRA_CA_CERTS file no longer breaks all connections
Readable certificates continue to be used when the certificate file is invalid, and a warning is printed to stderr.
Usage details
Who can use it: Connections using custom CA certificates.
How to use it: Set NODE_EXTRA_CA_CERTS and test the connection with 2.1.318.
Tip: Keep only valid PEM certificates in the variable.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesResponses are preserved when restarting after a large image or tool retry
Fixed an issue that dropped an interrupted response when restarting immediately after a large image or tool retry.
Usage details
Who can use it: Restartable sessions that use images and tools.
How to use it: Update to 2.1.318 and use the resume flow after an interruption.
Tip: Check that the session state has been saved before restarting.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesChainguard Harness added to Claude Marketplace
Chainguard Harness builds applications from source in isolated microVMs and probes running applications for exploitable vulnerabilities. It provides custom remediation recommendations for Claude’s findings.
Usage details
Who can use it: Partner product listed on Claude Marketplace; purchasing and access depend on the eligibility of the relevant Enterprise customer.
How to use it: Open the Chainguard Harness listing on Claude Marketplace and submit a purchase request, then define the application repository and test environment scope through partner setup.
Tip: Run your first trial on an isolated test application rather than in production, and verify recommended remediations by rescanning.
✓ Verified by official sourceSource: Claude Marketplace’te güvenlik, üretkenlik ve yaratıcı çalışma için yeni ajanlar ve ürünlerChannel polling cursors are now scoped per workspace
GitHub, GitLab, and DWS polling cursors are now stored separately for each workspace. Existing positions are preserved during upgrades, and channel state is stored with stricter permissions.
Usage details
Who can use it: Teams using GitHub, GitLab, or DWS Channel across multiple workspaces.
How to use it: Upgrade Qwen Code to v0.26.7; verify Channels separately in each workspace and check that new events are not fetched again after the upgrade.
Tip: Compare the latest event time in each workspace after upgrading; reprocessing old events may indicate a cursor scoping issue.
✓ Verified by official sourceSource: Qwen Code v0.26.7 sürüm notlarıChatGPT Images reuses saved reference photos in later chats
The new Saved reference images feature lets you name a photo and use it again in later chats without re-uploading it. For example, you can save your dog as Pepper and later request a watercolor portrait of @Pepper.
Usage details
Who can use it: Part of the ChatGPT Images 2.6 rollout; access depends on the account and phased rollout.
How to use it: Upload a photo in the Images workflow and save it under a name, such as Pepper. In a later image prompt, type @Pepper to use the same reference without re-uploading it.
Tip: Choose a single well-lit reference for the same character or product and give it a distinctive name that will be easy to find later.
ChatGPT starts generating looping GIFs and adjusting playback speed
With Images 2.6, ChatGPT can create looping GIFs from prompts; you can also upload a sketch and ask it to animate the character. You can adjust the generated GIF’s playback speed and share it.
Usage details
Who can use it: Rolling out to all plans on web, desktop, and mobile as part of the Images 2.6 rollout in ChatGPT.
How to use it: Ask ChatGPT to create a looping animation, or upload a sketch and ask it to animate the character. Once the result is generated, adjust the playback speed and share it with the Share option.
Tip: Explicitly describe the starting and ending poses, the sequence of movements, and that the loop should be seamless.
✓ Verified by official sourceModels: Images 2.6Source: ChatGPT sürüm notlarıSource: Images 2.6 tanıtımıCheckmarx One added to Claude Marketplace
Checkmarx One combines rule-based analysis with AI reasoning to find code vulnerabilities. It helps prioritize findings and fix them before code is released.
Usage details
Who can use it: Partner product listed on Claude Marketplace; enterprise purchasing and access terms vary by listing.
How to use it: Open the Checkmarx One listing on Claude Marketplace, submit a purchase request, and add the connection to your existing Checkmarx code analysis workflow using the partner guidelines.
Tip: Use rule-based findings alongside Claude’s prioritization and create a separate remediation queue for high-impact vulnerabilities.
✓ Verified by official sourceSource: Claude Marketplace’te güvenlik, üretkenlik ve yaratıcı çalışma için yeni ajanlar ve ürünlerClaude Tag admin setup fixes accessibility issues
Keyboard focus and screen reader announcements have been fixed; Escape no longer closes the Add dialog in the connect form.
Usage details
Who can use it: Teams setting up Claude Tag as an administrator.
How to use it: Open Admin setup and test the keyboard and screen reader flow.
Tip: Test Escape and screen reader announcements together.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesclaude.json restore command uses the correct backup
Fixed an incorrect backup reference and a read-only error in the restore command printed for a damaged or missing claude.json.
Usage details
Who can use it: Installations using configuration backups.
How to use it: Run the restore command provided at startup and verify the file.
Tip: Test backups before restoring them.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesThe Cloud MCP row shows the number of available tools
Cloud /mcp no longer counts tools the model can’t use and now shows the number available.
Usage details
Who can use it: Cloud sessions and MCP servers.
How to use it: Open the /mcp command in a cloud session and check the server row.
Tip: Distinguish the total number of tools from the number of active tools.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesCloud sessions can read GitHub Actions logs
gh run view --log and gh api log requests now return logs in most cloud environments instead of being redirected.
Usage details
Who can use it: Cloud sessions using GitHub Actions.
How to use it: Run gh run view --log or gh api in a cloud session.
Tip: Limit the output using specific run and job IDs.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesStandard Chrome names can be used in cloud sessions
In cloud sessions, the names google-chrome and CHROME_PATH let tools find the bundled Chromium.
Usage details
Who can use it: Tools that use Chromium in Cloud sessions.
How to use it: Call the browser through google-chrome or CHROME_PATH.
Tip: Try resolving CHROME_PATH first in the tool.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesCloud PR errors show the cause or a GitHub link
Create PR failures now provide the cause or a GitHub link instead of only showing Try again.
Usage details
Who can use it: Users creating PRs from Cloud and Remote Control.
How to use it: Try creating a PR and follow the error cause or link.
Tip: Don’t blindly retry the request without fixing the cause.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesThe Cloud Setup hook’s environment variables reach Bash
Fixed an issue where variables written by the Setup hook to CLAUDE_ENV_FILE did not reach Bash.
Usage details
Who can use it: Projects using the Setup hook in Cloud sessions.
How to use it: Write variables using the Setup hook and check them in the next Bash call.
Tip: Verify that secrets are present rather than logging them.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesCode Review analytics retries a failed request once
The Analytics page retries once after a network or server error, then displays a Try again banner.
Usage details
Who can use it: Teams using Code Review analytics.
How to use it: Open the Analytics page; if an error occurs, wait for the automatic retry.
Tip: Don’t conclude that data is missing before using the retry in the banner.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesActual image size limit shown for requests with many images
For more than 20 images, the error now shows the API’s actual limit instead of a fixed 2000px.
Usage details
Who can use it: Clients sending many images to the Claude API.
How to use it: Resend the request and resize the images based on the returned limit.
Tip: Check image dimensions on the client side in advance.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesCyera Agent Guardian added to Claude Marketplace
Cyera Agent Guardian inventories Claude deployments and manages permissions before Claude runs. It helps enforce policies for prompts, files, and tool calls, and test these controls with red-team exercises.
Usage details
Who can use it: Partner agent listed on Claude Marketplace; enterprise access and purchasing terms depend on the listing.
How to use it: Open the Cyera Agent Guardian listing on Claude Marketplace and submit a purchase request; define your deployments and the prompt, file, and tool policies to enforce during partner setup.
Tip: Before going live, run red-team scenarios with the most restrictive permission policies and measure false blocks.
✓ Verified by official sourceSource: Claude Marketplace’te güvenlik, üretkenlik ve yaratıcı çalışma için yeni ajanlar ve ürünlerDatadog Agent Observability added to Claude Marketplace
Datadog Agent Observability traces agent executions across tools and services. It surfaces production patterns and provides recommendations to improve agent quality, performance, and cost efficiency.
Usage details
Who can use it: Partner product listed on Claude Marketplace; enterprise purchasing and Datadog access are required.
How to use it: Open the Datadog Agent Observability listing on Claude Marketplace and submit a purchase request; connect Claude-based agent executions to the partner’s observability workflow.
Tip: Track success rate, latency, and cost per tool together on your first dashboard; don’t look only at total request volume.
✓ Verified by official sourceSource: Claude Marketplace’te güvenlik, üretkenlik ve yaratıcı çalışma için yeni ajanlar ve ürünlerDaybreak Red becomes available to approved customers through AWS GovCloud (US-East)
OpenAI expanded access to GPT-5.6 Cyber to the AWS GovCloud (US-East) environment through Daybreak Red. Authorized security teams can conduct vulnerability research, exploit validation, and security testing through Amazon Bedrock; access is subject to OpenAI approval and AWS readiness.
Usage details
Who can use it: Amazon Bedrock on AWS GovCloud (US-East) for approved government security teams and other eligible customers; Daybreak Red access approval is required.
How to use it: Complete the Daybreak application or discuss eligibility with your OpenAI/AWS account team. After approval, work with your AWS account team to enable Bedrock access; Daybreak Red approval alone does not grant model access.
Tip: Have an authorized security expert review the model’s findings and recommended fixes before putting them into production.
Delegation tools persist across profile switches, and goals can be resumed from the Web
A bug that caused delegation tools to disappear after switching profiles and returning was fixed, and delegation now respects the current Agent tool restrictions. Blocked goals can be resumed from the Web target bar; if the turn limit has been reached, increase it first.
Usage details
Who can use it: DeepSeek Harness Agent profiles, delegation tools, and the Web target bar.
How to use it: Switch profiles and return, then try the delegation tool again; resume the blocked goal from the Web target bar or increase the exhausted turn limit first.
Tip: Give subagents only the tools required for the goal, and set a realistic turn limit from the outset.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıExperimental Developer tools Statistics plugin adds a metrics view
The experimental Developer tools — Statistics plugin shows context, cache, model, and tool times by turn and step.
Usage details
Who can use it: Experimental developer tool enabled from the plugin manager.
How to use it: Enable the Developer tools — Statistics plugin in the plugin manager and open the turn or step metrics for the session.
Tip: When diagnosing latency, compare model, cache, and tool times separately instead of looking at the total time.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıdesign revoke no longer reports a 404 error as success
Fixed an issue where the command reported success when a /design revoke request returned 404.
Usage details
Who can use it: Users revoking Claude Design connections from the CLI.
How to use it: Run /design revoke with 2.1.318 and check the output.
Tip: Verify the server status after revoking.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesDesktop displays readable computer-use and tool-selection cards
Computer-use and tool-selection calls are now shown with clear titles, parameters, and result summaries. Users can see what each tool did without having to parse raw technical output.
Usage details
Who can use it: Kimi Code Desktop v1.3.0.
How to use it: When the Agent makes a computer-use or tool-selection call, open the card in the conversation and review its title, parameters, and result summary.
Tip: Before approving, check the target application, URL, and parameters to be entered in the card.
✓ Verified by official sourceSource: Kimi Code Desktop 1.3.0 resmi changelog’uDesktop adds diagnostics export and a startup recovery screen
Desktop now allows you to export diagnostics by packaging logs and crash reports in a single ZIP. When startup is slow or fails, the loading page also offers options to export diagnostics and restart.
Usage details
Who can use it: Kimi Code Desktop v1.3.0.
How to use it: Select `Help → Export Diagnostics…`; if the app won’t open, use the Export Diagnostics option on the loading screen.
Tip: Before sending the ZIP to the support team, check whether it contains source code, command output, or file paths.
✓ Verified by official sourceSource: Kimi Code Desktop 1.3.0 resmi changelog’uSource: Kimi Code Desktop güncel değişiklik özetiA new right-hand panel for session information is available in Desktop
The new right-hand panel brings together supporting information for the current session in one place. The panel stays in tabs alongside the session, and users can open or close details as needed.
Usage details
Who can use it: Kimi Code Desktop v1.3.0.
How to use it: Open the right-hand panel on the session screen and select the Session Information tab; expand the details if they’re collapsed by default.
Tip: In multi-directory projects, check the session context in the panel before taking action.
✓ Verified by official sourceSource: Kimi Code Desktop 1.3.0 resmi changelog’uSource: Kimi Code Desktop arayüz kılavuzuA saved login is no longer replaced with an empty file when storage is full
Fixed an issue where a saved login was replaced with an empty file when the network home directory ran out of space for data.
Usage details
Who can use it: Network home directories approaching their capacity limit.
How to use it: Free up quota and save the login again with 2.1.318.
Tip: Monitor home directory quotas regularly.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesTool selection errors fixed for four Claude-compatible models
Qwen Code fixed tool selection errors with Claude Opus 5.5, Claude Sonnet 5.5, Fable 5.1, and Mythos 5.1. Unsupported forced tool selection no longer causes failed calls with these models.
Usage details
Who can use it: Users running these models in tool-using workflows through Qwen Code.
How to use it: Select the relevant model and run a prompt that uses a tool; verify that the call completes with non-forced tool choice behavior.
Tip: When the model provider is updated, recheck the same tool call with a short regression prompt.
✓ Verified by official sourceModels: Claude Opus 5.5, Claude Sonnet 5.5, Fable 5.1, Mythos 5.1Source: Qwen Code v0.26.7 sürüm notlarıFile paths and Markdown links now open more accurately
Local tools expand a leading ~ or ~/ to the Host home directory; relative file links in chat use the session working directory at the time of the click. A bare ~ is now plain text in Markdown, and Windows backslashes and punctuation in bare URLs are also handled correctly.
Usage details
Who can use it: DeepSeek Harness local file tools, Markdown links, and Windows paths.
How to use it: Click a file link while in a worktree; try a path starting with ~/ and a Markdown link with Windows backslashes.
Tip: Because the session working directory can change, document file links with explicit absolute paths when sharing them long-term.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıA stopped response is no longer counted as a success because of a Mod
The turn.step hook can no longer finish a stopped response as successful in headless runs.
Usage details
Who can use it: Headless Agent SDK and claude -p Mod workflows.
How to use it: Upgrade to 2.1.318 and check the status of a stop scenario.
Tip: Handle stopping and success as separate states.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesFixed plugin ESM resolution and restricted sandbox launch issues
Plugin dependencies in configuration projects are now resolved using the project’s ESM rules. An issue that prevented restricted sandbox commands from launching when DSH ran with administrator privileges or in Windows Sandbox has been fixed; file write and delete restrictions remain in place.
Usage details
Who can use it: Configuration project plugins and Windows Sandbox or administrator-privileged installations.
How to use it: Launch the plugin configuration project using its own ESM rules; in Windows Sandbox, run a restricted command and separately check the file write and delete policies.
Tip: Running with administrator privileges does not remove sandbox security boundaries; still test write and delete permissions separately.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıPlugin manager adds rich details and third-party update support
The plugin manager can display rich-text details, and the page layout and navigation have been improved. Third-party plugin packages can be updated; if a hot update fails, the previous version is restored when possible.
Usage details
Who can use it: DeepSeek Harness plugin manager; third-party plugins and packages that support hot updates.
How to use it: Open the plugin manager, view a plugin’s details, and run the update action. If the hot update fails, check the restoration result.
Tip: Export your profile or save the release notes before updating third-party plugins.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıMissing working folder errors are now explained
Remote Control now reports the missing folder instead of showing a misleading spawn ENOENT error.
Usage details
Who can use it: Remote Control sessions whose working folder has been deleted.
How to use it: Re-create the specified folder or start with the correct folder.
Tip: Verify the remote folder before starting a session.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesNode-based Generated UI MCP runs in the Electron IDE companion
Qwen Code’s built-in Generated UI MCP server now runs properly in Node mode when launched from a CLI Electron-based IDE companion. This makes the Generated UI integration available through this client path.
Usage details
Who can use it: Generated UI users running Qwen Code with an Electron-based IDE companion.
How to use it: Launch Qwen Code CLI from the Electron IDE companion and test a tool call that produces MCP Apps/Generated UI output.
Tip: Verify the client’s MCP Apps display support beforehand; a running server alone does not mean visual cards will be displayed.
✓ Verified by official sourceSource: Qwen Code v0.26.7 sürüm notlarıPrevious thinking may not be preserved when resuming an old conversation
The behavior that preserved old thinking in 2.1.288 has been removed; thinking may be discarded in old conversations.
Usage details
Who can use it: Conversations started in 2.1.286 or earlier.
How to use it: Resume the old conversation and provide the necessary intermediate result again.
Tip: Don’t leave critical results only in the thinking field.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesGateway warning for federated Google token expiration
The gateway logs a warning when the Google token in the external_account credential file is nearing expiration.
Usage details
Who can use it: Gateway administrators using Google federation.
How to use it: Monitor gateway logs and renew the federation credential before the token expires.
Tip: Connect the warning to a centralized log alert.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesCopy history as Markdown and preserve replies in collapsed steps
The full chat history can now be copied as Markdown, with the original log path included when available. When work steps are collapsed, you can choose to preserve all assistant text or show only the final reply.
Usage details
Who can use it: DeepSeek Harness session view and work-step view.
How to use it: Copy the full history as Markdown from the session menu; when collapsing work steps, choose a view that preserves all replies or shows only the final answer.
Tip: When sharing a bug report, use the Markdown copy that includes the original log path; choose the final-answer view if you only need to read the result.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıVoice remixing API and AI Studio arrive for Gemini 3.8 Flash TTS
You can modify an existing prebuilt, Extended Voice Library, or designed voice with natural-language instructions to create a standalone, stored `voice_...` ID. The feature is available interactively in Google AI Studio and through `POST /v1beta/voices`; the generated `voice_...` ID can be used for synthesis with either of the two Gemini 3.8 TTS models.
Usage details
Who can use it: Projects on Gemini API Paid Tier 1 or above and Google AI Studio; project quota of 200 stateful voices per created voice, with storage for 1 year after last use.
How to use it: In Google AI Studio, open Generate Speech > Voice remixing; for the API, call `POST /v1beta/voices` with the fields `store: true`, `type: "prompted"`, `prompted.base_voice`, and `prompted.input`, then use the returned `voice_...` ID in a TTS request.
Tip: Change only one or two attributes (for example, pitch and pacing) with each remix; listen to the returned `sample_audio` preview before proceeding to the next remix.
✓ Verified by official sourceModels: gemini-3.8-flash-tts, gemini-3.8-flash-lite-ttsSource: Google AI Studio değişiklik günlüğü — 4 Kasım 2026Source: Voice remixing kılavuzuLogin expiration added to auth status and JSON output
The expiration date for a stored claude.ai login is provided as loginExpiresAt in auth status and as credential_expires_at in claude -p JSON.
Usage details
Who can use it: Claude Code CLI and claude -p JSON automations.
How to use it: Run claude auth status or read the field in the JSON output.
Tip: Set up monitoring to start the re-login flow before expiration.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesSSH fallback and prepare approval added for Git plugin installation
If HTTPS access to a self-hosted Git plugin fails but SSH access is available, installation switches to SSH. For plugins that include a prepare script, script permissions and retry options are shown.
Usage details
Who can use it: DeepSeek Harness plugin manager and self-hosted Git plugins.
How to use it: Install the Git plugin from the plugin manager; if HTTPS fails, have your SSH key ready, then grant permission at the prepare script prompt and choose to retry.
Tip: Test HTTPS and SSH access to self-hosted repositories in advance; grant permission to prepare scripts only when you trust their source.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıGitHub channel editor now covers repo, mention, and polling settings
The GitHub channel editor, which was previously made available, has been expanded; repository access, mention behavior, and notification polling settings can now also be edited from Web Shell. This reduces the need to configure these settings manually.
Usage details
Who can use it: Web Shell GitHub Channel users; the expanded scope of the previously available GitHub channel settings feature.
How to use it: In Web Shell, edit the GitHub repository access, mention behavior, and notification polling fields in the Channel editor.
Tip: Verify mention and polling settings in a small test repository to reduce notification noise.
✓ Verified by official sourceSource: Qwen Code v0.26.7 sürüm notlarıGlean added to Claude Marketplace
Glean brings together company knowledge and workflows across connected systems. It helps employees find answers from permitted sources and complete multistep work in connected applications.
Usage details
Who can use it: Partner product listed on Claude Marketplace; connected application permissions and enterprise purchasing terms apply.
How to use it: Open the Glean listing on Claude Marketplace and submit a purchase request; select the enterprise applications to connect and user access in partner administration.
Tip: For initial use, verify source permissions with a small user group and leave sensitive repositories out by default.
✓ Verified by official sourceSource: Claude Marketplace’te güvenlik, üretkenlik ve yaratıcı çalışma için yeni ajanlar ve ürünlerGlobal and project AGENTS.md budgets are now separate
Global and project AGENTS.md instructions now use separate maxBytes budgets; project instructions no longer displace global guidance. The combined message total from both chains can now reach up to twice the previous configured limit.
Usage details
Who can use it: DeepSeek Harness workspaces that use AGENTS.md.
How to use it: Keep global and project AGENTS.md files in their respective directories; review the existing maxBytes configuration and monitor the combined context from both instruction chains.
Tip: Keep important rules concise and prioritized; even with separate budgets, unnecessarily long instructions increase context size.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıImages upload before sending and the default limit is now 50 MiB
An image uploads as soon as it is added to the input field and can be reused when sending. The default single-image size limit has increased to 50 MiB.
Usage details
Who can use it: DeepSeek Harness sessions with image support; default 50 MiB limit per image.
How to use it: Drag or add an image to the input field, wait for the upload to complete, then send the message.
Tip: For large images, waiting for the preview to be ready before sending reduces the risk of uploading again.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıgrep -n options are correctly recognized in permission analysis
Shell permission analysis now recognizes `grep -n` and `grep --line-number` as options. This prevents grep calls intended to read files from incorrectly bypassing permission checks.
Usage details
Who can use it: All local and Managed sessions using Qwen Code shell tool permissions.
How to use it: Run a shell call using `grep -n` or `grep --line-number` and check that tool approval/permission analysis correctly classifies the file-reading behavior.
Tip: Rather than broadening shell permissions, review the tool description to confirm that the command is truly read-only.
✓ Verified by official sourceSource: Qwen Code v0.26.7 sürüm notlarıExperimental Auto review support added to headless mode
Experimental Auto review is now available in headless runs.
Usage details
Who can use it: DeepSeek Harness headless runs and the experimental Auto review feature.
How to use it: Enable Auto review in the headless profile and run a tool task to verify the review result in the logs.
Tip: Try headless automatic reviews with read-only tools first; add production write permissions later.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıImages can be attached by pasting an HTTP(S) URL
When a single HTTP(S) image URL is pasted into Web Shell, the image is now attached. Users also receive a warning when a file paste is unavailable.
Usage details
Who can use it: Desktop and touch clients using prompts with image attachments in Web Shell.
How to use it: Copy a single image HTTP(S) URL to the clipboard and paste it into the Web Shell composer; check that the attachment was added before sending.
Tip: Use a permanent, authorized image URL the model can access instead of a temporary or access-protected URL.
✓ Verified by official sourceSource: Qwen Code v0.26.7 sürüm notlarıThe IDE terminal no longer automatically installs and updates the extension
Claude Code launched from the IDE terminal no longer installs or updates the IDE extension; use /ide or the extension view instead.
Usage details
Who can use it: Users launching Claude Code from the IDE terminal.
How to use it: Run /ide in an external terminal or open the editor’s extension view.
Tip: Manage the extension version through centralized distribution.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesiPhone correctly shows when a cloud session turn ends
Fixed an issue where the iPhone app showed an ongoing cloud turn as finished.
Usage details
Who can use it: Those tracking cloud sessions with Claude for iOS.
How to use it: Update the iPhone app and monitor the cloud session.
Tip: Compare the notification status with the session in the app.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesThe Ask each time subagent doesn’t switch models for flagged messages
When model switching is set to Ask each time, the subagent stops instead of switching models.
Usage details
Who can use it: Sessions using a model policy for flagged requests.
How to use it: Set the option to Ask each time and run the flagged-message scenario.
Tip: Intentionally restart the stopped agent.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesKeys pressed early in permission prompts no longer cause incorrect responses
Fixed an issue where keys pressed before a slow-to-open permission prompt appeared could inadvertently answer it.
Usage details
Who can use it: Interactive terminal sessions.
How to use it: Update to 2.1.318 and use the normal keyboard flow when the prompt appears.
Tip: Don’t press the confirmation key before the prompt appears.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesSetup tokens from an allowed organization are no longer rejected
Fixed an issue where setup tokens belonging to an allowed organization were rejected on machines with forceLoginOrgUUID.
Usage details
Who can use it: Managed machines that enforce an organization ID.
How to use it: Configure the allowed organization’s token with setup-token and start it.
Tip: Store the token in an approved secure secrets store.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesClosed stdio MCP servers are no longer probed repeatedly
During version checks, closed stdio MCP servers are no longer retried repeatedly; a server that closes quickly uses the legacy handshake for seven days.
Usage details
Who can use it: Installations running stdio MCP servers.
How to use it: Start the MCP server with 2.1.318 and check /mcp status.
Tip: Don’t restart it without fixing the cause of the early shutdown.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesKimi Code displays backend model errors verbatim
The CLI now displays backend model error messages directly instead of reinterpreting them. Quota status is no longer inferred from the error text; explicit retryability information is required for retry policies.
Usage details
Who can use it: Kimi Code CLI 2.5.0 and custom provider/model integrations.
How to use it: When a model call fails, handle the original backend message and retry status directly from the CLI output.
Tip: Apply automatic retries only to errors explicitly marked as retryable.
✓ Verified by official sourceSource: Kimi Code CLI 2.5.0 resmi changelog’uSource: Kimi Code Server API referansıKimi Code Desktop now requires macOS 13 or later
With Desktop v1.3.0, macOS 13 is now the minimum supported version; macOS 12 can no longer receive new updates. Windows and x86_64 Linux packages continue to be available.
Usage details
Who can use it: Kimi Code Desktop v1.3.0; macOS 13 and later, Windows, x86_64 Linux.
How to use it: Check your macOS version in About This Mac; if you’re on macOS 12, upgrade the operating system or stay on a compatible older version before installing v1.3.0.
Tip: For enterprise Mac deployments, plan the operating system upgrade before updating Desktop.
✓ Verified by official sourceSource: Kimi Code Desktop 1.3.0 resmi changelog’uSource: Kimi Code Desktop hızlı başlangıç kılavuzuKimi Code for VS Code automatically refreshes the model catalog
The VS Code integration has been updated to automatically refresh the provider model catalog. This reduces the need to manually reimport catalog changes when a provider adds a new model.
Usage details
Who can use it: Kimi Code for VS Code users with the Kimi Code 2.5.0 canary 2026-11-04 change.
How to use it: Update Kimi Code for VS Code and reload the VS Code window; the list refreshes automatically when the provider or model catalog is opened.
Tip: After automatic refresh, check the model’s context and capability information before using it in a production request.
✓ Verified by official sourceSource: Kimi Code 2.5.0 canary resmi GitHub sürümüSource: Kimi Code for VS Code hızlı başlangıç kılavuzuKimi Code disables global full-text search by default
Kimi Code no longer creates a global full-text index by default to scan past sessions in the background. Web search is therefore unavailable by default, but can be re-enabled with an explicit setting.
Usage details
Who can use it: Kimi Code CLI/Web 2.5.0; users who want to search all past sessions.
How to use it: Set `[database] global_search = true` in `config.toml` or provide the `KIMI_CODE_GLOBAL_SEARCH=1` variable.
Tip: Keep the default disabled behavior for sensitive projects; enable search only for trusted local sessions when needed.
✓ Verified by official sourceSource: Kimi Code CLI 2.5.0 resmi changelog’uSource: Kimi Code yapılandırma referansıKimi Code can change the current Goal objective within a session
Users can edit the objective text of an ongoing Goal in place without starting a new session. This preserves existing progress and session context when the objective’s scope changes.
Usage details
Who can use it: Kimi Code CLI 2.5.0 canary 2026-11-04 and versions that include this change.
How to use it: Replace the active objective with a new one using `/goal replace <objective>`; the Goal’s status and progress remain in the same session.
Tip: When changing the objective, include a measurable completion condition in the new text.
✓ Verified by official sourceSource: Kimi Code 2.5.0 canary resmi GitHub sürümüSource: Kimi Code slash komutları referansıKimi Code subagent model selection and `secondary_model.force` behavior changed
`secondary_model.force` is now ignored and triggers a warning; implicit `primary` selection has been removed for single-model pools. Model selection for Agent/AgentSwarm spawns is available again when two or more aliases are defined.
Usage details
Who can use it: Kimi Code CLI 2.5.0; users of custom subagent model pools.
How to use it: List at least two model aliases under `[secondary_model.models]` in `config.toml`; use only `default_model` for a fixed model and do not add `force`.
Tip: Define a fast model and a powerful model with descriptive aliases to distinguish the cost and speed of subtasks.
✓ Verified by official sourceSource: Kimi Code CLI 2.5.0 resmi changelog’uSource: Kimi Code yapılandırma referansıWeb Search and Fetch tools are unavailable without credentials
The model’s Web search tools are hidden if the required credentials aren’t available; the new Web API-based Fetch tools are also made available based on credential readiness.
Usage details
Who can use it: DeepSeek Harness profiles that use Web Search and Web Fetch and have credentials configured.
How to use it: Enter the required API credentials in Settings → Web Search and verify the connection; then check the tool list in a new session.
Tip: Check the settings validation first to distinguish missing credentials from a lack of tool support in the model.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıSwitching to a smaller-context model and prefix caching are safer
Before switching to a model with a smaller context window, history is compressed using the old model; if compression fails, the switch is halted and the previous request header is preserved. If the system prompt has not changed, new requests reuse the existing context prefix cache.
Usage details
Who can use it: DeepSeek Harness sessions that use model switching and session context.
How to use it: Select a model with a smaller context window from the model selector and wait for compression to finish; allow the cache to be reused for consecutive requests without changing the system prompt.
Tip: In long sessions, switch models at a point that leaves time for compression rather than immediately before a critical step.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıUser steering can immediately stop waiting for background output
When a user instruction arrives while waiting for background task output, the wait ends immediately and returns the current output; the background task continues running.
Usage details
Who can use it: DeepSeek Harness background tasks and user steering.
How to use it: Send an instruction while a background task is running; receive the current output and check that the background task continues without stopping.
Tip: If you don’t want to stop a long-running task, send the new request as steering rather than using the cancel action.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıRequesting more usage from the usage limit menu takes one step
The Ask your admin for more usage option now submits the request in one step.
Usage details
Who can use it: Plans with administrator-approved usage limits.
How to use it: Select Ask your admin for more usage in the usage limit menu.
Tip: Add brief task context to the request.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesA finished agent no longer runs again after a usage limit
Fixed an issue where workflows resumed after hitting a limit reran finished agents.
Usage details
Who can use it: Sessions using workflows and agent chains.
How to use it: Resume the workflow that hit the limit with 2.1.318.
Tip: Design agent tasks to be idempotent.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesQueued Desktop messages use the session’s current permission mode
Messages queued while the Agent is running now use the session’s permission mode when they’re processed, rather than the setting that was in effect when they entered the queue. This keeps behavior consistent if the user changes the permission mode and runs a pending message later.
Usage details
Who can use it: Kimi Code Desktop v1.3.0.
How to use it: Send a message while the Agent is running, then change the permission mode below Composer and wait for the message to be processed from the queue.
Tip: When side-effecting operations are queued, explicitly check Always Ask before choosing Ask When Needed for automated operations.
✓ Verified by official sourceSource: Kimi Code Desktop 1.3.0 resmi changelog’uSource: Kimi Code Desktop görev ve izin kılavuzuGit operations are more reliable with the built-in Bash on macOS
Reliability of creating Git commits and pull requests in macOS’s /bin/bash shell has been improved.
Usage details
Who can use it: Git workflows using the built-in Bash on macOS.
How to use it: Upgrade to 2.1.318 and run the commit and PR workflow.
Tip: Verify the diff and remote branch before opening a PR.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesMultiple Bash sessions no longer stall on macOS
Memory growth and stalls have been fixed in sessions running many Bash commands on macOS.
Usage details
Who can use it: Sessions running multiple Bash commands on macOS.
How to use it: Upgrade Claude Code to 2.1.318 and rerun the workflow.
Tip: Monitor memory during long-running parallel commands.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesManaged Web Shell entry and session error notifications improved
A Managed host entry has been added to Web Shell. The Managed empty state layout has been improved, and session failure and unconfirmed submission notifications are now grouped above the composer.
Usage details
Who can use it: Managed Web Shell users.
How to use it: Open the Managed host entry on the Web Shell home page; if the session doesn’t load, follow the error or verification notification above the composer.
Tip: If a session isn’t added, check the unconfirmed submission warning above the composer before trying again.
✓ Verified by official sourceSource: Qwen Code v0.26.7 sürüm notlarıPhone OTP options for lead forms added in Marketing API v27
With Graph API v27.0, lead forms can use the phone_otp_verification_type field: NOT_ENABLED, OTP_REQUIRED, or OTP_OPTIONAL. OTP_OPTIONAL is available only for eligible ad accounts, and an authorized ad_account_id must be provided when creating a form; the legacy is_phone_sms_verify_enabled field also remains supported.
Usage details
Who can use it: Graph API v27.0+; lead form reading and creation flows; OTP_OPTIONAL is available only for eligible ad accounts.
How to use it: Read the value with GET /{lead-gen-form-id}?fields=phone_otp_verification_type; send phone_otp_verification_type in a POST /{page-id}/leadgen_forms request, and include an eligible ad_account_id for OTP_OPTIONAL.
Tip: When reading older forms, handle null values and the legacy field together in your client.
✓ Verified by official sourceSource: Graph API ve Marketing API v27.0 ChangelogSource: Graph API v27.0 ve Marketing API v27.0 duyurusuProduct showcase can be enabled automatically in Marketing API v27
If the Product showcase setting is absent from a creative, Meta can automatically enable the product_extensions feature when it finds a high-confidence match with a specific product set in the catalog. Because the setting remains on the creative, it also affects subsequent ads that use the same creative.
Usage details
Who can use it: Marketing API v27.0+; requests to create new ads and ad creatives with high-confidence catalog matches.
How to use it: If you don’t want automatic enrollment, send degrees_of_freedom_spec.creative_features_spec.product_extensions.enroll_status=OPT_OUT; if you prefer, check the recommendation before creation using the catalog_suggestions_preview endpoint.
Tip: If you want full control over brand or product selection, review the preview result and explicitly set OPT_OUT.
✓ Verified by official sourceSource: Graph API ve Marketing API v27.0 ChangelogSource: Graph API v27.0 ve Marketing API v27.0 duyurusuDiagnostic report added for desktop startup failures
For desktop startup failures, a diagnostic report can be obtained that includes a limited history of recent plugin operations.
Usage details
Who can use it: DeepSeek Harness desktop app; users experiencing startup errors.
How to use it: When the desktop app fails to start, open the diagnostic report option and attach the plugin operation history from the report to your support request or bug report.
Tip: Before sharing the report, check that it does not contain an API key or sensitive file paths.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıMCP add and remove operations write configuration under a lock
claude mcp add and remove now wait for the configuration lock; if the lock cannot be acquired, they return exit code 1.
Usage details
Who can use it: Processes accessing MCP configuration concurrently.
How to use it: Run the MCP command; if the lock is busy, wait or handle the error code.
Tip: Serialize jobs that write to the same configuration in parallel.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesThe tool catalog refreshes in the same turn when MCP permissions or tools change
After MCP OAuth is completed or the tool configuration changes, Kimi Code refreshes the available tools within the current turn. There’s no need to wait for a new session or manually restart.
Usage details
Who can use it: Kimi Code CLI 2.5.0 and sessions using an MCP server.
How to use it: Authorize or modify the MCP server with `/mcp-config`; expect Agent to use the new tool catalog in the same turn.
Tip: After authorization, check that Agent has refreshed its tool list before starting a new turn.
✓ Verified by official sourceSource: Kimi Code CLI 2.5.0 resmi changelog’uSource: Kimi Code Plugins kılavuzuMessages and drafts are now finalized only after Host acceptance
A message isn’t displayed and the draft isn’t cleared until it’s accepted by Host. The input field is read-only during submission; if an error occurs, the draft, quote, and attachments are preserved.
Usage details
Who can use it: DeepSeek Harness Web and desktop session input field.
How to use it: Send a message and wait for Host acceptance. If submission fails, retry with the draft, quote, and attachments in the input field.
Tip: When working with large attachments, don’t close the window without checking that the content was preserved after a submission error.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıMod activation responses are now recorded
When Claude Code asks to activate Mods it has written, it adds your response to the session record.
Usage details
Who can use it: All sessions using Claude Code 2.1.318.
How to use it: Update Claude Code to 2.1.318 and respond when the Mod prompt appears.
Tip: Check the session record to audit Mod behavior later.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesMod fork tasks can have a separate effort level
The $.model.fork call in Mod can now accept effort; the subtask runs at a level independent of the main session and reports that level.
Usage details
Who can use it: Claude Code users developing Mods.
How to use it: Add, for example, effort: low to the $.model.fork call.
Tip: Run short, low-risk subtasks with low.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesStop works on the message immediately after a model change
In headless sessions, the message immediately after a model change can now be stopped or canceled with Stop.
Usage details
Who can use it: Headless Claude Code sessions.
How to use it: Change the model and use Stop on the running message.
Tip: Wait for the model switch before sending a new request.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesNew settings for model catalog refresh and usage telemetry
Qwen Code adds the operator-wide `model.catalogRefresh` setting. When `privacy.usageStatisticsEnabled:false` is used, background model catalog refreshes stop, but local model metadata remains available.
Usage details
Who can use it: Operators using Qwen Code v0.26.7 and installations managing privacy settings.
How to use it: Set `model.catalogRefresh` in the operator configuration; use `privacy.usageStatisticsEnabled:false` to disable background catalog refreshes.
Tip: When disabling catalog refreshes, make sure the local catalog data is up to date; otherwise, new model names may not appear.
✓ Verified by official sourceSource: Qwen Code v0.26.7 sürüm notlarıImproved model picker and provider-specific model ordering
Configured models are selected by default when the model picker opens; adding and removing candidates, and selecting all visible candidates, are supported. Models from the DeepSeek and pi-ai providers can be reordered by dragging or using the arrow keys.
Usage details
Who can use it: DeepSeek Harness model picker; DeepSeek and pi-ai provider configurations.
How to use it: In Settings → Models, select or remove candidates; drag a model in the DeepSeek or pi-ai provider list, or use the arrow keys, then save.
Tip: Move your most-used model to the top of the list to reduce the chance of selecting the wrong provider and avoid unnecessary clicks.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarımonday agents added to Claude Marketplace
monday agents uses context from projects, boards, documents, and meetings, along with owners and due dates. It gives PMO, IT, marketing, finance, and operations teams a way to control AI access, approvals, and spending.
Usage details
Who can use it: Partner agents listed on Claude Marketplace; requires a monday workspace and enterprise permissions configuration.
How to use it: Open the monday agents listing on Claude Marketplace and submit a purchase request; define the boards to be used, ownership rules, and approval workflows on the partner side.
Tip: Initially keep agents’ spending and approval permissions in read-only or human-approval mode.
✓ Verified by official sourceSource: Claude Marketplace’te güvenlik, üretkenlik ve yaratıcı çalışma için yeni ajanlar ve ürünlermonday.com becomes available as a Connected App in Gemini
monday.com is now available as a Connected App in Gemini for personal, work, and school Google accounts; users can ask about information on their boards in a Gemini chat using `@monday.com`. On Google Workspace, the integration is also available through Marketplace for working with monday data in Gmail, Docs, Slides, and Sheets.
Usage details
Who can use it: Gemini web app, mobile app, and Windows app; English; countries and regions where Gemini and monday.com are supported; personal, work, or school Google account.
How to use it: Connect monday.com from Gemini Settings > Connected Apps, or type `@monday.com` in any Gemini chat and select the app; Workspace users can also install the monday.com integration from Google Workspace Marketplace.
Tip: Narrow your first query to a specific board and time range (for example, overdue tasks); check the connected monday workspace’s access scope before using the response.
✓ Verified by official sourceSource: monday.com — Google Gemini’de kullanılabilir olduSource: Google Gemini Connected Apps uygunluk tablosuNetlify Enterprise added to Claude Marketplace
Netlify Enterprise deploys applications, sites, and agents built with Claude to its global edge network. It provides preview URLs for every change, along with SSO, role-based access, and audit logs.
Usage details
Who can use it: Partner product listed on Claude Marketplace; Enterprise governance and purchase terms apply.
How to use it: Open the Netlify Enterprise listing on Claude Marketplace and submit a purchase request; connect a Claude output to a Netlify project to start a preview deployment.
Tip: Review every change at its preview URL before sending it to production, and keep publishing permissions limited.
✓ Verified by official sourceSource: Claude Marketplace’te güvenlik, üretkenlik ve yaratıcı çalışma için yeni ajanlar ve ürünlerNode PTC adds VM Script mode and optional state persistence
Node PTC now includes VM Script mode, which supports top-level await, dynamic imports, and expression results. State persistence across calls is optional; by default, execution resets on every call.
Usage details
Who can use it: DeepSeek Harness plugins and SDK flows that use Node PTC.
How to use it: Enable VM Script mode in the Node PTC configuration; also enable state persistence if you need to share variables and imports across calls.
Tip: Enable state persistence only in trusted, short-lived sessions; the default behavior of independent calls is more isolated.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notları262.144-token context is preserved for OpenRouter qwen3-coder:exacto
The `qwen/qwen3-coder:exacto` identifier on OpenRouter now preserves the base model name and is recognized with a 262.144-token context window instead of 200K. This ensures the model’s actual context capacity is correctly passed to Qwen Code.
Usage details
Who can use it: Users of OpenRouter `qwen/qwen3-coder:exacto` through Qwen Code.
How to use it: Select OpenRouter as the model provider and use `qwen/qwen3-coder:exacto` as the model identifier; verify the capacity with a long-context prompt.
Tip: Before filling the context window, also check the provider’s actual limits and pricing.
Validation of specially named marketplaces is now consistent
plugin validate no longer reports success for a specially named file that marketplace add would reject.
Usage details
Who can use it: Developers managing plugin marketplaces.
How to use it: Run claude plugin validate first, then marketplace add.
Tip: Use an accepted format for the marketplace filename.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesCustom plugin market migrated to an index-based flow
Custom plugin market has been renamed Custom plugin index; the discovery page reads user-added index files and excludes custom/private releases. Official bundle lists are exported from the dsh-official-bundles package.
Usage details
Who can use it: DeepSeek Harness plugin discovery and custom plugin indexes; integration developers.
How to use it: Use Custom plugin index in the plugin discovery settings, add the user index file, and check that custom releases aren’t listed; for custom app-boot integrations, switch to the dsh-official-bundles source.
Tip: List only published, trusted releases in custom indexes; don’t include packages tagged private in the distribution chain.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıCustom model capacity and step model routing have changed
For custom models with unspecified capacity, the default context is set to 1M and the maximum output to 128K. Agent now fixes the step model before preparing the prompt and before the pre-step; the same model is retained for retries, and plugins must use agent.model to select a model.
Usage details
Who can use it: Custom OpenAI-compatible endpoints and Agent plugins.
How to use it: Explicitly configure context and output capacity for a custom model with lower capacity; use agent.model to select a model from a plugin, and don’t switch routes with agent/request.
Tip: If your endpoint doesn’t support 1M/128K, explicitly set a lower capacity instead of relying on delayed compaction.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıInvalid `max_tokens` error fixed for custom model registries
The output token limit declared by models retrieved from a custom registry is now saved. This fixes the 400 `Invalid max_tokens` error seen with some models from custom registries.
Usage details
Who can use it: Kimi Code CLI 2.5.0 and models imported from a custom registry.
How to use it: Re-import the registry with `kimi provider add <url>` or refresh the provider model list; requests will use the saved output limit.
Tip: Match the model output limits in the registry’s `api.json` to the actual service limits.
✓ Verified by official sourceSource: Kimi Code CLI 2.5.0 resmi GitHub sürümüSource: Kimi Code komut referansıCustom slash commands produce output in p and SDK sessions
Fixed an issue where custom slash commands that failed or were called only by Claude were silent in p and SDK sessions.
Usage details
Who can use it: Users of custom slash commands with claude -p or Agent SDK.
How to use it: Run the command with 2.1.318 and read the text or SDK output.
Tip: In automation, distinguish empty output from error output.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesAI-generated disclosure management added to Page post editing
With Graph API v27.0, apps can add or remove AI-generated self-disclosure when editing existing Page posts. Setting is_ai_generated=true adds the label, while false removes it; if the field is omitted, the current value is retained.
Usage details
Who can use it: Graph API v27.0+; POST /{post-id} and POST /{page-post-id}; no changes in older versions.
How to use it: Send the boolean is_ai_generated field in the post edit request; true adds the label, false removes it, and omitting the field retains the current value.
Tip: If you don’t need to change the label, omit the field to preserve the existing disclosure value.
✓ Verified by official sourceSource: Graph API ve Marketing API v27.0 ChangelogSource: Graph API v27.0 ve Marketing API v27.0 duyurusuUpgraded to pi-ai 1.3.0; malformed tool parameters are rejected early
The model catalog, endpoints, and reasoning replay rules have been updated with pi-ai 1.3.0. If completed tool parameters are malformed, the response is terminated before any tool runs.
Usage details
Who can use it: DeepSeek Harness plugins and providers that use pi-ai routes.
How to use it: Update the profile that uses pi-ai and verify that tool schemas produce valid JSON parameters; reload the model catalog after the update.
Tip: Validate tool parameters against the schema before production so early terminations don’t disrupt the user flow.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıSlow token refresh in plugin commands is fixed
Fixed an issue where a plugin command exited before token refresh finished, causing the session to be dropped on the next run.
Usage details
Who can use it: Claude Code CLI installations using plugins.
How to use it: Update to 2.1.318 and run the plugin command again.
Tip: Wait for token refresh to finish before shutting down.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesPR review report now shows final cohort and endpoint details
PR review routing now selects the cohort before the model and reports the final bucket, cohort, model, and endpoint hostname. This makes it clear which route canary or experimental routing took.
Usage details
Who can use it: Teams using the Qwen Code PR review workflow.
How to use it: Run a PR review; check the bucket, cohort, model, and endpoint hostname fields in the completed report.
Tip: When model quality changes, compare the cohort and endpoint in the report first to distinguish routing differences.
✓ Verified by official sourceSource: Qwen Code v0.26.7 sürüm notlarıprompt.submit hooks are now available in the Ink terminal
Mods `prompt.submit` hooks now run in the Ink terminal before the typed prompt is processed. Hooks can see the composer text and expanded pasted content; this extends the previous headless support to the terminal.
Usage details
Who can use it: Qwen Code Mods developers using the Ink terminal.
How to use it: Define a `prompt.submit` hook in a Mod and verify in the Ink terminal that it observes or transforms the text before the prompt is submitted.
Tip: For hooks that transform text, check the logs to see which form of the text is sent to the model after pasted content is expanded.
✓ Verified by official sourceSource: Qwen Code v0.26.7 sürüm notlarıprompt.submit Mod content stays in the user turn
Content added at the prompt.submit stage stays in the user turn instead of becoming a system message.
Usage details
Who can use it: Users developing Mods and hooks.
How to use it: Run the Mod with 2.1.318 and inspect the prompt.submit content.
Tip: Configure persistent system rules separately.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesRalph loop and Session search removed from the experimental list
Ralph loop and Session search are no longer shown in the experimental plugins list. This indicates that the related experimental list and discovery path have been removed.
Usage details
Who can use it: DeepSeek Harness experimental plugin list and profile discovery.
How to use it: Open the experimental list in the plugin manager and check that Ralph loop and Session search are no longer listed; update profiles that use them according to the migration notes.
Tip: Export profiles that depend on removed experimental items and identify an alternative workflow before upgrading.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıMissing value for release-idle-session-min triggers a warning
A release-idle-session-min argument with no value or only whitespace now triggers a startup warning.
Usage details
Who can use it: Self-hosted runner administrators.
How to use it: Provide a numeric value for the argument and check the startup log.
Tip: Pass arguments as a shell array.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesRetention cleanup preserves the latest good backup
When cleanup runs only in the home directory used by claude -p, it takes a fresh backup before deleting the latest good backup.
Usage details
Who can use it: Non-interactive installations using retention cleanup.
How to use it: Update to 2.1.318 and check the backups after cleanup.
Tip: Keep an external backup in addition to automatic retention.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesRHEL 10 EPEL bootstrap RPM pinned
The EPEL bootstrap RPM used in the Qwen Code Desktop installation flow on RHEL 10 has been pinned. This fix improves installation reliability for the RHEL 10 RPM distribution added on November 2.
Usage details
Who can use it: Users of RHEL 10 x64 Desktop packages.
How to use it: Update Qwen Code Desktop to v0.26.7 or later and rerun the RHEL 10 RPM installation.
Tip: Before changing the EPEL RPM separately in your enterprise mirror, verify that the release uses the pinned package.
✓ Verified by official sourceSource: Qwen Code v0.26.7 sürüm notlarıRubrik Code Guardian added to Claude Marketplace
Rubrik Code Guardian examines a secure copy of GitHub or Azure DevOps repositories for multi-step attack paths. It ranks issues by severity and helps turn verified critical findings into tracked tickets.
Usage details
Who can use it: Partner agent listed on Claude Marketplace; subject to repository access, purchase, and enterprise authorization terms.
How to use it: Open the Rubrik Code Guardian listing on Claude Marketplace, submit a purchase request, and select the GitHub or Azure DevOps repositories to scan in the partner connection flow.
Tip: First scan a read-only repository copy, then enable the ticketing rule for critical findings with team approval.
✓ Verified by official sourceSource: Claude Marketplace’te güvenlik, üretkenlik ve yaratıcı çalışma için yeni ajanlar ve ürünlerRequest reason added to the runner spawn hook
CLAUDE_RUNNER_SPAWN_REASON adds the reason, such as runner_lost or resumed, to the runner hook environment.
Usage details
Who can use it: Organizations using the self-hosted runner spawn-runner hook.
How to use it: Read the CLAUDE_RUNNER_SPAWN_REASON variable in the hook.
Tip: Route empty or unknown values to a safe default path.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesRunway added to Claude Marketplace
Runway creates and edits video, images, and audio using natural language. It helps creative and marketing teams turn a brief into finished media assets.
Usage details
Who can use it: Partner product listed on Claude Marketplace; access and purchase terms depend on Runway and the enterprise agreement.
How to use it: Open the Runway listing on Claude Marketplace and submit a purchase request; then provide a creative brief in natural language to start a video, image, or audio generation workflow.
Tip: Include the target format, duration, brand constraints, and revision criteria in the brief from the start.
✓ Verified by official sourceSource: Claude Marketplace’te güvenlik, üretkenlik ve yaratıcı çalışma için yeni ajanlar ve ürünlersdk-minimal keeps recovery records for short sessions
sdk-minimal writes a session recovery record before a model request or tool run. If the record cannot be written, the operation does not proceed; this makes it safe to resume interrupted short sessions.
Usage details
Who can use it: DeepSeek Harness sdk-minimal configuration and sessions created with the SDK.
How to use it: Start a session with sdk-minimal, configure storage where the recovery record can be written, then run a model or tool call.
Tip: Store recovery records in persistent, backed-up storage rather than a temporary directory.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıLong Search API summaries are prioritized for the model
Instead of the short excerpts on cards, the Search function prioritizes long Search API summaries of up to 800 characters per source by default for the model.
Usage details
Who can use it: DeepSeek Harness searches using Search API; the default source summary limit is 800 characters.
How to use it: Enable Web search and run a search task; inspect the long source summaries in the model context separately from the short search card shown to users.
Tip: For research involving multiple sources, preserve the 800-character summaries and use only verified claims in the final answer.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıSession Projection and storage compatibility changes
The Session Projection registration API is now asynchronous, and plugins must retain a handle at startup. JSON and SQLite storage can read older versions explicitly declared as compatible and write in the current format.
Usage details
Who can use it: Session Projection plugins and self-hosted installations using JSON/SQLite.
How to use it: Wait for the asynchronous Projection registration result when initializing the plugin and store the handle; explicitly specify a compatible older version for storage.
Tip: Back up the database before upgrading storage, and enable only documented compatibility versions.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıA new cloud session can be opened from a deleted environment card
Deleted cloud environment cards now show a New session button.
Usage details
Who can use it: Users who see a deleted cloud environment card.
How to use it: Click the New session button on the card and select a new environment.
Tip: Select the repository and environment settings again.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesDeleted transcript is no longer recreated from the last prompt
Fixed an issue where a transcript deleted when an idle, non-interactive session closed was recreated from the last prompt.
Usage details
Who can use it: Non-interactive or claude -p-based sessions.
How to use it: Update to 2.1.318 and reapply transcript cleanup.
Tip: Clearly define your transcript retention policy.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesSubagent responses can wait longer with a cold cache
In Bedrock, Vertex, Foundry, or sessions with telemetry disabled, pending responses can wait for up to five minutes.
Usage details
Who can use it: The specified providers and sessions with telemetry disabled.
How to use it: Start the subagent and wait up to five minutes for a progress summary.
Tip: Don’t cancel early when the cache is cold.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesAdd chat text to the input field as a quote
Text from chat can now be transferred to the input field as a quote. After submission, the quote is displayed as a brief preview.
Usage details
Who can use it: Chat input field in the DeepSeek Harness web and desktop interfaces.
How to use it: Select the relevant text in a chat message and transfer it to the input field as a quote; check the compact preview before sending.
Tip: In code reviews, quote only the necessary lines to reduce context and token usage.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıThe subagent_oar plugin supports local Codex and Claude CLI tasks
The experimental subagent_oar tool plugin adds the ability to delegate independent tasks to local Codex or Claude CLI processes.
Usage details
Who can use it: Available as an experimental plugin in DeepSeek Harness profiles; on machines with local Codex or Claude CLI installed.
How to use it: Enable the experimental subagent_oar plugin in the plugin manager, configure local CLI credentials, and delegate the independent task to this tool.
Tip: Keep workspace permissions as narrow as possible before granting local CLIs access to files and commands.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıThe right renewal option is recommended for expired logins
An expired login is clearly reported; claude -p auth login and setup-token are recommended.
Usage details
Who can use it: Users whose saved claude.ai login has expired.
How to use it: Run the recommended claude auth login or claude setup-token command.
Tip: Monitor token expiration for CI.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesclaude -p warning for an expiring login
A stderr warning is added to claude -p text output 10 days before a stored claude.ai login expires.
Usage details
Who can use it: CLI and CI workflows that use claude -p text output.
How to use it: Upgrade to 2.1.318 and monitor the stderr stream as well.
Tip: Collect stdout and stderr in separate channels.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesVersion cleanup no longer deletes another user’s process
Fixed an issue where a running Claude Code process belonging to another user was mistaken for a dead process and its binary was deleted.
Usage details
Who can use it: Shared machines and processes belonging to different users.
How to use it: Upgrade to 2.1.318 and run version cleanup again.
Tip: If possible, use the same user for updates and running the process.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesCommitted Anthropic spend can be used for Marketplace solutions
Eligible Enterprise customers can direct part of their committed Anthropic spend to partner solutions on Claude Marketplace. This makes it possible to purchase Claude and partner products under a single commitment.
Usage details
Who can use it: For solutions on Claude Marketplace; eligible Enterprise customers. Eligibility and commitment scope are determined by Anthropic.
How to use it: Open the Agents & products section on Claude Marketplace, select a listing, and submit a purchase request; confirm with your Anthropic account team whether your committed balance can be applied.
Tip: Before ordering, confirm in writing which partner products are covered by your commitment.
✓ Verified by official sourceSource: Claude Marketplace’te güvenlik, üretkenlik ve yaratıcı çalışma için yeni ajanlar ve ürünlerTanium Cloud added to Claude Marketplace
Tanium Cloud provides real-time visibility and control over endpoints. It helps teams find and fix vulnerabilities, misconfigurations, and missing patches.
Usage details
Who can use it: Partner product listed on Claude Marketplace; access and purchase depend on the relevant enterprise agreement.
How to use it: Open the Tanium Cloud listing on Claude Marketplace and submit a purchase request; configure the endpoint inventory and patch management scope in the partner account.
Tip: First monitor critical asset groups under a separate policy, then gradually expand the scope of automated remediation.
✓ Verified by official sourceSource: Claude Marketplace’te güvenlik, üretkenlik ve yaratıcı çalışma için yeni ajanlar ve ürünlerDelivery cards now include file references from the entire turn
File links in delivery cards and closing responses now include deliveries and writes made after the response. It’s clarified that Agent must provide the file link before waiting for a response from the user.
Usage details
Who can use it: DeepSeek Harness file deliveries and closing responses.
How to use it: Create or write a file during a turn, then open the delivery card and the file link in the closing response to check the final file.
Tip: In workflows that wait for user approval, provide the file link first; don’t leave it until the next turn.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıTrajectory preserves actual output from failed tools and legacy code highlighting
The actual text output from failed tools is preserved in the Trajectory list, details, and copied content. TypeScript and Python syntax highlighting is restored for affected legacy run_code records.
Usage details
Who can use it: DeepSeek Harness Trajectory view and historical run_code records.
How to use it: After a failed tool call, open the Trajectory details and copy output; select the language used in a legacy run_code record and check the highlighting.
Tip: When debugging, rely on the tool’s actual return text, not a generated error summary.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıTrajectory adds a raw JSON view for live tool results
The Trajectory results page can show the raw JSON of the live tool response above the processed view.
Usage details
Who can use it: DeepSeek Harness Trajectory results page.
How to use it: After a tool call completes, open the Trajectory results page and expand the raw JSON view above the processed result.
Tip: When diagnosing an integration error, compare raw JSON field names instead of processed text.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıSuccess status corrected for turn.step interruptions
Responses interrupted by a turn.step API error are no longer reported as successful in claude -p and the Agent SDK.
Usage details
Who can use it: Agent SDK and claude -p Mods that use turn.step.
How to use it: Upgrade to 2.1.318 and check the exit code for an API error scenario.
Tip: In CI, rely on the exit code rather than the text.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesunifiedAgentLoop expanded to ACP scheduled and background turns
When `experimental.unifiedAgentLoop` is enabled, ACP cron turns and background notification turns also use the shared loop. This preserves shared turn, tool grouping, and stop behavior in these turns.
Usage details
Who can use it: ACP clients running scheduled tasks or background notifications; scope expansion in v0.26.7.
How to use it: Enable `experimental.unifiedAgentLoop`; start a cron or background-notification turn through ACP.
Tip: To prevent duplicate notifications in background turns, compare the results of the old and new loops with the same client.
✓ Verified by official sourceSource: Qwen Code v0.26.7 sürüm notlarıunifiedAgentLoop now also covers ACP Goal turns
With the experimental `unifiedAgentLoop` enabled, ACP Goal turns now use the turns, tool groups, and stop points in the shared loop. This expands the previously documented ACP shared-loop coverage to the Goal flow.
Usage details
Who can use it: ACP clients using `experimental.unifiedAgentLoop`; Qwen Code v0.26.7.
How to use it: Enable `experimental.unifiedAgentLoop` in Qwen Code experimental settings and run the ACP Goal flow.
Tip: Before putting the Goal flow into production, test that tool groups and stop point behavior are compatible with your client.
✓ Verified by official sourceSource: Qwen Code v0.26.7 sürüm notlarıRemote Control reconnects faster after sleep or suspension
The reconnection speed of remote sessions after sleep or suspension has been improved.
Usage details
Who can use it: Remote Control and remote session users.
How to use it: Open the session after sleep and observe the reconnection.
Tip: Verify the connection status before the first critical command.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesRemote file reading, writing, searching, and completion are faster
Network round trips have been reduced for file reading, editing, writing, searching, and completion in remote environment sessions. This change reduces interaction latency, especially in SSH- and container-based projects.
Usage details
Who can use it: Kimi Code CLI 2.5.0 remote environment sessions.
How to use it: Define a remote environment and start a session with `kimi --environment <id>`; no additional settings are required.
Tip: Referencing a relevant folder as a whole instead of many small files saves even more network calls.
✓ Verified by official sourceSource: Kimi Code CLI 2.5.0 resmi GitHub sürümüSource: Remote environments kılavuzuRemote access errors no longer expose internal details to callers
The remote access error-handling flow no longer displays internal error details to callers. Various remote access reliability and security issues were also fixed in the same release.
Usage details
Who can use it: Clients and service owners that use `qwen serve` remote access.
How to use it: After enabling remote access, review errors on the client side; for detailed diagnostics, check the server logs as an authorized operator.
Tip: Keep safe errors returned to users and authorized server logs in separate channels.
✓ Verified by official sourceSource: Qwen Code v0.26.7 sürüm notlarıWebSocket compression is now enabled by default for remote connections
WebSocket compression is enabled by default to reduce the transfer of large histories over remote connections. ETag validation for unchanged web static assets also prevents them from being downloaded again with a 304 response.
Usage details
Who can use it: Remote DeepSeek Harness connections; compression can be disabled with the websocketCompression setting.
How to use it: Start the remote connection as usual; to disable compression, set websocketCompression to false in the configuration.
Tip: Keep default compression enabled on slow networks; on CPU-constrained local networks, consider disabling it only after measuring.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıA prolonged upload rejection no longer stops a cloud turn
Fixed an issue that caused a cloud session to stop mid-turn when the service rejected uploads for several minutes.
Usage details
Who can use it: Claude Code users using Cloud sessions.
How to use it: Upgrade to 2.1.318 and rerun the task that requires an upload.
Tip: For large files, monitor the task and service status together.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesVS Code sign-in screen shows CLI error output
When the sign-in status check fails, the VS Code screen shows Claude Code’s explanation.
Usage details
Who can use it: Users whose sign-in check fails in VS Code.
How to use it: Update the extension to 2.1.318 and follow the explanation on the sign-in screen.
Tip: Forward the exact error text to your administrator.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesVS Code no longer enters a process loop when sign-in is rejected by a setting
Fixed an issue where VS Code started a new Claude Code process every minute when an organization setting rejected sign-in.
Usage details
Who can use it: Organization-managed sign-in settings in VS Code.
How to use it: Update the extension to 2.1.318 and reopen the window.
Tip: Share the enterprise sign-in policy with users.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesWebflow added to Claude Marketplace
Webflow provides built-in tools for creating, managing, and optimizing websites. It allows marketers, designers, developers, and agents to make changes with governance controls.
Usage details
Who can use it: Partner product listed on Claude Marketplace; enterprise access and purchase terms apply.
How to use it: Open the Webflow listing on Claude Marketplace and submit a purchase request; add your workspace to the Claude workflow using the partner connection.
Tip: Test change approvals and role limits in a draft workspace before granting publishing permissions.
✓ Verified by official sourceSource: Claude Marketplace’te güvenlik, üretkenlik ve yaratıcı çalışma için yeni ajanlar ve ürünlerRollout begins for separate, nameable chats for Meta AI in WhatsApp
WABetaInfo and SammyFans report that in WhatsApp Android beta 2.26.44.6, Meta AI threads are being turned into separate conversations in the chat list. Users can give a new chat a name and custom profile picture, and find these chats using the AI filter in the Chats tab; there is no public official confirmation from Meta or general release date.
Usage details
Who can use it: WhatsApp beta for Android 2.26.44.6; some beta testers, phased rollout.
How to use it: Open a Meta AI conversation, tap the new chat option in the three-dot menu, and set a name and profile picture; open the conversation using the AI filter in the Chats tab.
Tip: Use a separate chat for each topic to reduce context mixing and the need to search for old threads.
Desktop update flow improved on Windows and macOS
Windows installer progress and window interactions have been fixed; the app publisher and names in the uninstall list are now displayed correctly. macOS updates now handle exiting full screen, workspace recovery after a failed installation, and in-window notifications that match the app theme more effectively.
Usage details
Who can use it: DeepSeek Harness desktop app; Windows and macOS.
How to use it: Use the in-window flow when an update notification appears in the desktop app; on macOS, exit full screen before confirming the update and use the retry option if it fails.
Tip: Save the status of open tasks before a desktop update, and make sure the update window is visible in full-screen apps.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıSource: DeepSeek Harness masaüstü indirme sayfasıWorkspace file uploads can now run concurrently
Workspace uploads can now proceed at the same time. The system continues to reject occupied destinations and enforce directory capacity protections.
Usage details
Who can use it: Managed Agent users using Qwen Code Workspace.
How to use it: Submit multiple files at the same time on the Workspace upload screen; rejection behavior for occupied destinations is unchanged.
Tip: When uploading many files, check destination directory names in advance; conflicting destinations will still be rejected.
✓ Verified by official sourceSource: Qwen Code v0.26.7 sürüm notlarıOption to bypass the workspace trust prompt with `require_trust = false` added
Kimi Code CLI now offers a configuration option to bypass workspace trust prompts. When `[workspace] require_trust = false` is set, new workspaces don’t require trust approval.
Usage details
Who can use it: Kimi Code CLI 2.5.0; trusted automation environments.
How to use it: Add `[workspace]` and `require_trust = false` to `~/.kimi-code/config.toml`, then start a new session.
Tip: Enable this setting only for folders whose contents and execution permissions are controlled.
✓ Verified by official sourceSource: Kimi Code CLI 2.5.0 resmi changelog’uSource: Kimi Code yapılandırma referansıOn restart, an older message with an API error is no longer moved ahead
Fixed an issue where an older message that received an API error was resent before a new message after a restart.
Usage details
Who can use it: CLI, cloud, and runner sessions that can be restarted.
How to use it: Upgrade to 2.1.318 and restart the interrupted session.
Tip: Record message IDs in workflows where ordering matters.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesMCP tool returns its schema when called before loading
The error response for an MCP call made before loading now includes the tool’s input schema.
Usage details
Who can use it: Clients that call MCP tools programmatically.
How to use it: Test the scenario where a tool is called before loading, then read the error schema.
Tip: Manage loading state and don’t retry blindly.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesZoom Agentic Search added to Claude Marketplace
Zoom Agentic Search searches meeting summaries, workplace conversations, documents, and connected third-party sources from within Claude. It provides permission-aware results and works without analyzing raw transcripts.
Usage details
Who can use it: Partner agent listed on Claude Marketplace; Zoom and connected-source permissions, as well as enterprise purchase terms, apply.
How to use it: Open the Zoom Agentic Search listing on Claude Marketplace and submit a purchase request; select the Zoom workspace and permitted third-party sources in the partner connection.
Tip: Before using results for critical decisions, verify the permissions and freshness of the source meeting or document.
✓ Verified by official sourceSource: Claude Marketplace’te güvenlik, üretkenlik ve yaratıcı çalışma için yeni ajanlar ve ürünlerLarge tool results are shown as condensed previews in ACP session replays
The live ACP tool result view now shows large file-editing results as saved session previews instead of full old and new content. This reduces unnecessarily long output in session replays.
Usage details
Who can use it: Users of ACP session replays or Managed Web Shell history.
How to use it: After a large file-editing turn, open the ACP session replay view; inspect the saved preview instead of the full result.
Tip: When you need the full diff, open the saved session or artifact linked from the preview; don’t use the replay screen as a raw diff repository.
✓ Verified by official sourceSource: Qwen Code v0.26.7 sürüm notlarıSix new interface languages added
Text and localization configurations have been added for Japanese, Spanish, French, German, Russian, and Brazilian Portuguese.
Usage details
Who can use it: DeepSeek Harness interface; Web and desktop clients with the relevant localization configuration.
How to use it: Choose one of the supported languages from the language selector in Settings, then reopen the interface to apply the localization.
Tip: For team use, agree on a shared language to avoid confusion over screenshots and menu names.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıRuntime duration updates live in background shell details
Fixed an issue where the runtime duration of a background shell in an open dialog did not update.
Usage details
Who can use it: Sessions running background shell commands.
How to use it: Send a shell command to the background and keep its details dialog open.
Tip: Use the live duration as a helpful indicator, not as a completion signal.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesMulti-agent research workflow documented for AutoClaw Agent Cluster
Z.ai’s official post dated November 4 details how AutoClaw Agent Cluster plans complex research, breaks it into independent tasks, runs them with parallel agents, and reviews the source-supported results. The post does not say that this feature was first released in a new version; it should be treated as a new official usage workflow, not as a change.
Usage details
Who can use it: Desktop users of AutoClaw Agent Cluster; the post specifies no plan, country, or regional restrictions.
How to use it: Select Agent Cluster in the conversation input field; add the product list or source files; specify the research period, source types, comparison areas, and deliverable format in the same prompt.
Tip: Give each research branch the same fields (capabilities, access, pricing, evidence, and uncertainty) so the results can be compared.
✓ Verified by official sourceSource: AutoClaw ile çok ajanlı araştırmaWorkspaces and sessions with the same name are easier to distinguish in the sidebar
When selecting workspaces with the same name, the shortest distinguishing path suffix is shown. The working directory is now visible on session cards, and sidebar search matches session ID fragments regardless of case.
Usage details
Who can use it: DeepSeek Harness workspace selector, sidebar, and session search.
How to use it: Open workspaces with the same name in the selector; hover over a session in the sidebar or type part of a session ID in the search box.
Tip: Adding the last few characters of a session ID to a team note makes it easier to find sessions with similar names.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıConnection downtime and Trajectory message order fixed
The chat timer pauses when the connection is lost or being re-established, then resumes upon reconnection. After a stop, subsequent routing and follow-up messages are preserved in Trajectory in log order, before the assistant output.
Usage details
Who can use it: DeepSeek Harness web and desktop sessions; Trajectory history.
How to use it: Disconnect and reconnect an active session, then check the timer; review follow-up messages after stopping in Trajectory.
Tip: For long tasks with network interruptions, use the interface timer—not wall-clock time—to measure actual working time.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıPan large images and deliver original files by dragging
Images that exceed the viewport can now be panned by dragging. Original files can be delivered using in-app or desktop drag and drop.
Usage details
Who can use it: DeepSeek Harness Web and desktop image previews.
How to use it: Drag an image that exceeds the viewport in the preview; drag and drop the original file into the app or onto the desktop.
Tip: For analysis, deliver the original file by drag and drop instead of using a reduced preview.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıclaude -p and the SDK no longer write unchanged config
Non-interactive claude -p and SDK runs no longer rewrite claude.json if it hasn’t changed.
Usage details
Who can use it: claude -p and SDK automations.
How to use it: Upgrade to 2.1.318 and compare the file’s timestamp.
Tip: Trigger file monitoring only when there’s an actual change.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesConnector search indicates a partial directory
Connector results tell Claude that the search covers only part of the directory.
Usage details
Who can use it: Users of Cloud sessions and connector search.
How to use it: Search, then narrow or repeat the search based on the partial-results information.
Tip: Don’t treat a single result as evidence that the entire directory is empty.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesExperimental personalized plugin recommendations added
The experimental personalized recommendations plugin can show cards under relevant responses for enabling an official plugin or creating a personalized page.
Usage details
Who can use it: DeepSeek Harness users who enable the experimental Personalized recommendations plugin.
How to use it: Enable the Personalized recommendations plugin in the plugin manager; use the enable or page-creation cards under relevant responses.
Tip: Review which plugins are trusted before enabling recommendation cards as a team standard.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıDesktop File Tree can expand symbolic link directories
File Tree can now open directories represented by symbolic links and browse the files inside them. This broadens file discovery in projects that use monorepos and shared source folders.
Usage details
Who can use it: Kimi Code Desktop v1.3.0.
How to use it: Open the File Tree panel and expand a directory marked with the symbolic link icon.
Tip: Open only the links you need to avoid unnecessary scanning of circular symbolic links.
✓ Verified by official sourceSource: Kimi Code Desktop 1.3.0 resmi changelog’uDesktop opens Markdown files in the default system app
Kimi Code Desktop v1.3.0 can open Markdown documents in the default system app as well as in its in-app preview. This makes it easier to quickly review a document in an external editor or Markdown reader.
Usage details
Who can use it: Kimi Code Desktop v1.3.0; macOS, Windows, and Linux desktop users.
How to use it: Open a Markdown file in File Preview and select `Open` from the file actions.
Tip: Use Open and Preview in sequence to compare the source Markdown with the rendered preview.
✓ Verified by official sourceSource: Kimi Code Desktop 1.3.0 resmi changelog’uSource: Kimi Code görev yürütme kılavuzuDesktop adds a Follow default model option to subagent model settings
Subagent model settings now include the `Follow default model` option, which follows the default model in the main configuration. This means you don’t need to manually sync the subtask setting when the main model changes.
Usage details
Who can use it: Kimi Code Desktop v1.3.0.
How to use it: Open the subagent model settings in Settings and select `Follow default model`.
Tip: Use a fixed model for fast, cost-controlled subtasks and Follow default model for tasks that vary.
✓ Verified by official sourceSource: Kimi Code Desktop 1.3.0 resmi changelog’uFile search and Word delivery work more efficiently
File searches with a fixed directory prefix are faster because irrelevant directories are skipped. Routine Word edits are delivered after content and formatting are saved and verified, without unnecessary ZIP, XML, or metadata inspection.
Usage details
Who can use it: DeepSeek Harness file tools and Word tasks.
How to use it: Start a file search with a specific directory prefix; after editing a Word document, check the save and formatting verification results, then use delivery.
Tip: In large repositories, limit searches to the working directory or a fixed prefix whenever possible.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıFile tab position and dark theme settings navigation fixed
The scroll position is preserved in sidebar file tabs after selecting, opening, or closing a tab. The hover background for settings navigation is now visible in dark theme.
Usage details
Who can use it: DeepSeek Harness sidebar file tabs and dark theme settings.
How to use it: Open multiple file tabs in the sidebar and close one; then hover over Settings navigation in dark theme.
Tip: When reviewing multiple files, use preserved scroll position while closing tabs to avoid searching the file list again.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıInvalid additional workspace paths are now shown in error messages
When starting the TUI, an invalid additional workspace directory no longer causes a silent exit; the problematic path is shown to the user. This makes it easier to identify which folder caused a configuration error in multi-directory projects.
Usage details
Who can use it: Kimi Code CLI 2.5.0 and users of multi-directory workspaces.
How to use it: Start the CLI; if an error occurs, correct the displayed directory path and try again with `config.toml` or the session’s additional directory list.
Tip: Specify additional directories using absolute paths that exist; remove outdated paths after moving directories.
✓ Verified by official sourceSource: Kimi Code CLI 2.5.0 resmi GitHub sürümüGemini CLI nightly fixes extension, model, hook, and sandbox bugs
The November 4 nightly release fixes streaming errors in extension downloads, preserves an explicitly selected `gemini-2.5-flash` model during dynamic model resolution, and fixes `SubagentStop` hook matching. It also improves timely replay of ACP session/load history and `--userns=keep-id` support in rootless Podman.
Usage details
Who can use it: Gemini CLI pre-release nightly channel; release tag `v0.67.0-nightly.20261104.g89e9d25d6`.
How to use it: Upgrade your nightly installation to `v0.67.0-nightly.20261104.g89e9d25d6`; then rerun extension release download, hook, ACP session/load, and rootless Podman workflows.
Tip: After upgrading the nightly build, teams using CI/CD or Podman sandbox should run a brief smoke test to confirm that previous model selections and hook configurations are preserved.
✓ Verified by official sourceModels: gemini-2.5-flashSource: Gemini CLI nightly 4 Kasım 2026 sürüm notuSwitching between teams in the Gemini Enterprise mobile app reaches GA
The Gemini Enterprise mobile app can now switch between multiple teams without signing out and scanning a QR code again. The first team opened is saved to the list; users can rename or remove a team from that list.
Usage details
Who can use it: Gemini Enterprise mobile app; documented as a GA feature for Standard, Plus, Pay-as-you-go, and Frontline editions.
How to use it: Sign in to the Google Gemini Enterprise mobile app with your company account; after the initial connection, select another team from the teams list, or rename or remove a team using the list item’s options.
Tip: Administrators working across multiple organizational environments can reduce QR code and sign-in steps by saving each team with a meaningful name.
✓ Verified by official sourceSource: Google Cloud sürüm notları — 4 Kasım 2026Source: Gemini Enterprise mobil uygulamasını kullanmaInput and session statistics expanded
Truncated statistics in the input field can now be viewed in full by hovering over them. Session statistics also show the number of turns and steps.
Usage details
Who can use it: DeepSeek Harness input field and session statistics panels.
How to use it: Hover over a truncated statistic; open session statistics to review the number of turns and steps.
Tip: When assessing token usage or tool duration in long tasks, record the number of turns and steps at the same time.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıHarness admission, recovery, and Hook checks are faster
Standalone Harness admission, recovery, cleanup, and Hook admission checks no longer use the tenant placement lock in safe cases. Managed Agent performance improves while decision outcomes remain unchanged.
Usage details
Who can use it: Teams using Standalone Harness and Managed Agent infrastructure.
How to use it: Upgrade to v0.26.7; run an admission, recovery, or cleanup flow and check that the verdict behavior is unchanged.
Tip: In recovery tests, verify that both the duration and verdict remain the same as in the previous version.
✓ Verified by official sourceSource: Qwen Code v0.26.7 sürüm notlarıKimi Code removes Edit and `sed` restrictions for incremental edits
The rules requiring the Edit tool for incremental file edits and prohibiting Bash `sed` have been removed. The Agent can use different editing methods when appropriate.
Usage details
Who can use it: Kimi Code CLI 2.5.0 and later.
How to use it: No additional settings are required; assign a file-editing task and review the Edit or Bash approach chosen by the Agent in the approval flow.
Tip: Even though the restriction has been removed, review the diff for large changes and prefer small, reversible steps.
✓ Verified by official sourceSource: Kimi Code CLI 2.5.0 resmi changelog’uKimi Code shows todo update reminders less often
The Agent now receives a reminder if it hasn’t updated the todo list for 10 steps instead of 5, and avoids unnecessary warnings when the list still matches progress. This reduces repeated planning messages on long tasks.
Usage details
Who can use it: Kimi Code CLI 2.5.0.
How to use it: No additional setup is needed; start a long task and track updates in `/tasks` or the todo card.
Tip: For long Goal tasks, manually updating the todo list at key milestones is more reliable than waiting for an automatic reminder.
✓ Verified by official sourceSource: Kimi Code CLI 2.5.0 resmi changelog’uUsage credit notifications no longer keep repeating
Fixed a bug that showed the credit notification every 30 seconds while a background subagent was running.
Usage details
Who can use it: Plans using background subagents and usage credits.
How to use it: Update to 2.1.318 and run a background subagent.
Tip: Check actual usage with /usage.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesNative macOS folder picker available on first launch
Fixed an issue that caused the native macOS folder picker to appear inactive on first launch and reject folder clicks.
Usage details
Who can use it: DeepSeek Harness macOS desktop app.
How to use it: When adding a workspace, open the folder picker, click a folder, and confirm that the selection is accepted.
Tip: If the folder does not appear on first launch, verify that the latest candidate build is installed instead of restarting the app.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıManaged Runtime no longer displays internal instructions
The `record_artifact` directive has been removed from Managed Runtime worker results. The result returned to the user no longer carries this internal recording instruction unnecessarily.
Usage details
Who can use it: Qwen Code users who use Managed Runtime.
How to use it: Run a Managed Runtime task and use the result text directly. If artifact recording is needed, follow the product’s actual artifact flow separately.
Tip: Check the Managed Runtime recording list instead of assuming an artifact was recorded based on the result text.
✓ Verified by official sourceSource: Qwen Code v0.26.7 sürüm notlarıManaged turn submission and migration preparation involve less lock contention
Admission and migration preparation for Managed Agent turn submission have been decoupled from the tenant placement lock where safe. This reduces wait times without changing result behavior.
Usage details
Who can use it: Deployments that use Managed Agent turn and migration operations.
How to use it: Upgrade to v0.26.7 and run the turn submission and migration preparation flows in the same session; monitor operation wait times.
Tip: Measure error rates and wait times together during migration; don’t treat a shorter queue alone as a sign of success.
✓ Verified by official sourceSource: Qwen Code v0.26.7 sürüm notlarıCanceling MCP OAuth no longer interrupts other authorization attempts
Canceling an MCP OAuth authorization no longer accidentally stops another authorization attempt that is pending at the same time. Authorization flows work more independently for teams setting up multiple MCP servers.
Usage details
Who can use it: Kimi Code CLI 2.5.0 and multi-server setups using MCP OAuth.
How to use it: Authorize multiple servers through `/mcp-config`; when you cancel one flow, check that the other continues.
Tip: Use separate authorization windows and clear names for each server to avoid confusing which OAuth flow you’re canceling.
✓ Verified by official sourceSource: Kimi Code CLI 2.5.0 resmi changelog’uSource: Kimi Code Plugins kılavuzuMonitor and workflow notifications are attributed to the correct turn
Monitor and workflow completion notifications are now attributed to the turn that recorded them. This lets the active turn’s results reach the next model request sooner.
Usage details
Who can use it: Sessions that use Qwen Code monitor and workflow.
How to use it: Run a monitor or workflow; observe the model request and result generated in the turn that follows the one that recorded the completion notification.
Tip: When running many background tasks at once, check the session logs to see which turn each notification came from.
✓ Verified by official sourceSource: Qwen Code v0.26.7 sürüm notlarıRegular conversation branches are no longer counted as subagents
Regular conversation branches run independently and are no longer mistakenly classified as subagents; the creator relationship is preserved in actual subagent sessions.
Usage details
Who can use it: DeepSeek Harness session branches and actual subagent sessions.
How to use it: Create a regular branch from a session and observe it running independently; when you create a subagent, check that the parent session relationship is preserved.
Tip: Use regular branches for independent experiments, and subagents for delegated tasks and linked work.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıSession commands respond with fewer lockups
Lock usage has been reduced for common Session commands such as changing the model, approving, context compression, canceling tasks, and renaming. This aims to make the same session respond faster.
Usage details
Who can use it: Teams using Qwen Code Managed Agent Session.
How to use it: Upgrade to v0.26.7 and change the model, approve, start context compression, cancel a task, or rename it in Session.
Tip: In busy sessions, record the time between command submission and the UI response to measure user-perceived latency.
✓ Verified by official sourceSource: Qwen Code v0.26.7 sürüm notlarıpermissions views moved to the standard dialog
permissions details and delete confirmation now use the standard dialog; pressing Ctrl+C twice returns or cancels.
Usage details
Who can use it: Interactive Claude Code sessions.
How to use it: Open /permissions and use the rule details or delete view.
Tip: Check the dialog context when canceling.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesDuplicate PushNotification notifications are fixed
Fixed a duplicate notification that occurred when the status field was omitted from a PushNotification call.
Usage details
Who can use it: Sessions and integrations using PushNotification.
How to use it: Update to 2.1.318 and run the scenario without the status field.
Tip: Deduplicate notifications by ID in the consumer.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesResident Runtime Session acquisition and release are faster
Managed Agent resident Runtime Session claim and release flows no longer wait for the tenant placement lock. Session placement and release wait times have been reduced while preserving successful behavior.
Usage details
Who can use it: Managed Agent deployments that use Resident Runtime Session.
How to use it: Update Managed Agent to v0.26.7 and run the resident Runtime Session flow with the same concurrency.
Tip: When measuring performance gains, record not only the first response but also the total claim-to-release time.
✓ Verified by official sourceSource: Qwen Code v0.26.7 sürüm notlarıRight sidebar tabs can be closed with the middle mouse button
Right sidebar tabs can be closed with the middle mouse button; closing a background tab does not change the current selection, and the initial tab is preserved.
Usage details
Who can use it: The right sidebar in the DeepSeek Harness desktop and Web interfaces.
How to use it: Middle-click the tab you want to close in the right sidebar; the initial tab cannot be closed.
Tip: When many files or results tabs are open, use the middle mouse button to quickly close only the unnecessary tabs.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıCompleted automation tasks can be cleared with a single confirmation
The automation tasks page now includes an option to clear all inactive tasks with a single confirmation while preserving running tasks.
Usage details
Who can use it: DeepSeek Harness automation tasks page.
How to use it: Open the automation tasks page, select the action to clear completed or failed tasks, and confirm; ongoing tasks are preserved.
Tip: Export the output of failed tasks before clearing; the action removes only inactive tasks.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıInteractions in question and approval cards clarified
For multiple-choice questions, all recommended options are selected by default; long question and option descriptions can be expanded and collapsed inline. Approval messages clearly indicate whether permission is actually being requested from the user.
Usage details
Who can use it: DeepSeek Harness question cards, tasks with options, and approval flows.
How to use it: Open a multiple-choice card and check that recommendations are selected; expand a long description and read the permission wording in the approval message before proceeding.
Tip: Review automatically selected options individually before submitting; a recommended selection does not imply permission.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıSteer messages appear dimmed until the Agent reads them
Steer messages sent to a running Agent now appear dimmed until the Agent reads them, then return to normal. The status carries over to subsequent turns, making it easier to tell whether a message was actually processed.
Usage details
Who can use it: Kimi Code CLI 2.5.0.
How to use it: Send a message with `Ctrl-S` while the Agent is running, and watch the dimmed/read status on the message card.
Tip: For critical directions, don’t pile on new requests until you confirm the Agent has applied the message in its next response.
✓ Verified by official sourceSource: Kimi Code CLI 2.5.0 resmi changelog’uSource: Kimi Code etkileşim kılavuzuTerminal fork names better distinguish conflicts
The second terminal fork is named Name (fork 2) when the app doesn't assign it a name.
Usage details
Who can use it: Users who fork with Remote Control.
How to use it: Create a terminal fork, fork it again, and check the name.
Tip: Distinguish sessions with meaningful names.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesOverwritten local Markdown images now display the latest version
After a local Markdown image is overwritten, new messages and newly opened previews no longer use the stale image from the cache.
Usage details
Who can use it: DeepSeek Harness local Markdown image previews.
How to use it: Replace the local Markdown image, then open a new message or preview; verify that the latest file is displayed.
Tip: In image-generation workflows, open a new preview after each change to avoid accidentally using a stale cached image.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıA long error word no longer pushes the close button out of view
Fixed an issue where a long error word in the VS Code Claude panel hid the close button.
Usage details
Who can use it: Claude Code VS Code extension users.
How to use it: Update the extension to 2.1.318 and open the error banner.
Tip: If needed, copy the error text before closing it.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesWeb Fetch can read pages with mixed UTF-8 declarations
Web Fetch can now read valid pages whose encoding declaration contains multiple labels alongside UTF-8. This eliminates unnecessary read errors on pages with multiple encoding declarations.
Usage details
Who can use it: Web Fetch tool and web pages serving valid UTF-8 content.
How to use it: Enable Web Fetch and request the URL of a page with mixed encoding declarations; check that the returned text is not corrupted.
Tip: If the encoding issue persists, also verify the page’s raw response headers and content encoding.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıWorkspace operations use the tenant placement lock less often
Normal workspace execution, member removal, and file rewind operations no longer acquire the tenant-wide placement lock. This reduces unnecessary blocking across workspaces.
Usage details
Who can use it: Teams running multiple Managed Agent workspaces within a tenant.
How to use it: Upgrade Qwen Code Managed Agent to v0.26.7 and use concurrent workspace execution, member removal, or file rewind in the normal workflow.
Tip: For busy tenants, compare wait times before and after the upgrade using the same workload.
✓ Verified by official sourceSource: Qwen Code v0.26.7 sürüm notlarıSide-by-side checkouts are distinguished in the environment menu
Checkouts of the same repository appear in the Remote Control menu under their own folder names.
Usage details
Who can use it: Users with multiple Remote Control checkouts for one repository.
How to use it: Open the environment menu and compare the folder names.
Tip: Keep folder names short and distinctive.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesAnswered question groups remain in chat bubbles
Answered question groups are displayed as message bubbles. Responses to Native and PTC calls remain visible even when work steps are collapsed.
Usage details
Who can use it: DeepSeek Harness sessions and question-and-answer tool calls.
How to use it: Answer when a tool call asks a question, then collapse the work steps and check that the answered question group remains in a chat bubble.
Tip: Keep the session record instead of deleting question bubbles to preserve decision history in multi-step tasks.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-rc.1 resmî release notlarıHelp notifications say to try again instead of using a terminal-only key
You should know notifications say try again instead of press 2, which does not work in the GUI.
Usage details
Who can use it: Claude Code interfaces outside the terminal.
How to use it: Open the notification and use the try again action.
Tip: Select the button shown in the non-terminal interface.
✓ Verified by official sourceSource: Claude Code v2.1.318 release notesThe local FetchURL tool can access private network addresses
Kimi Code Agent’s local FetchURL tool has been expanded to access private addresses. This makes it possible to test local development services or internal endpoints through Agent’s own FetchURL flow.
Usage details
Who can use it: Versions of Kimi Code CLI 2.5.0 that include the 2026-11-04 canary change; network access depends on the machine’s permissions.
How to use it: Ask Agent to inspect an accessible local or private network URL; the FetchURL tool is used if network and security policies allow it.
Tip: Limit private network access to trusted URLs; don’t expose credentials or administrative endpoints to Agent.
✓ Verified by official sourceSource: Kimi Code 2.5.0 canary resmi GitHub sürümüLocal Workspace read, upload, and status checks are faster
Local Workspace file read, upload, and status checks reuse CLI capability probes while preserving the same local identity. This reduces the cost of repeated local checks.
Usage details
Who can use it: Managed Agent setups using Workspace with a local CLI identity.
How to use it: Upgrade to v0.26.7; run consecutive file read, upload, and status checks with the same CLI identity.
Tip: Remember that the probe cache must be refreshed when the identity or capabilities change; don’t compare performance using stale results.
✓ Verified by official sourceSource: Qwen Code v0.26.7 sürüm notlarıAgent SDK hooks can block a flow when a callback fails
The `onFailure: "block"` option was added to Agent SDK hook callbacks. When a callback fails, the relevant operation can now be blocked instead of continuing.
Usage details
Who can use it: Developers using Agent SDK hook callbacks.
How to use it: Add `onFailure: "block"` to the hook definition and update the SDK in flows where you want the action to stop when the callback fails.
Tip: Use `block` for security or compliance checks; for temporary telemetry hooks, keep the existing behavior to avoid unintended interruptions.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesAgentScope Java supports the JEV structured decision model
JEV integration has been added to AgentScope Java through `agentscope-extensions-jev`. Developers can use `JevClient` and review components to add structured decisions to tool selection, model routing, execution guardrails, evidence processing, and trajectory evaluation workflows.
Usage details
Who can use it: AgentScope developers using Java 17+; JEV API access and separate JEV credentials are required.
How to use it: Add the `io.agentscope:agentscope-extensions-jev` dependency to the project; set `TYPESAFE_API_KEY` and run the first decision through `JevClient` and `SystemOneRequest`.
Tip: Don’t interpret JEV probability as universal truth; calibrate thresholds using your own business data and route uncertain results to human review.
✓ Verified by official sourceModels: jev-latestSource: Teaching Agents to Decide: From JEV Integration to the Full Harness Decision FlowSubtask permissions retain the same restrictions as the main session
Subtask permissions now match the main session’s permissions. In Plan Mode, subtasks remain subject to the same restrictions.
Usage details
Who can use it: ZCode v3.17.2 and later; users who use subtasks or multiple agents.
How to use it: Start a multi-agent task in Plan Mode and check that subtask permissions match the main session’s.
Tip: Before starting parallel subtasks, keep the main session’s access scope to the minimum necessary.
✓ Verified by official sourceSource: ZCode Releases & Updates — v3.17.2Goal Mode introduced in AutoClaw for multi-step tasks
AutoClaw Goal Mode takes a verifiable output and completion criteria from the user, then plans, executes, and repeatedly evaluates the task. If evidence or conditions are missing, it can start another work cycle.
Usage details
Who can use it: AutoClaw users; available in a new conversation. No plan or country restrictions specified.
How to use it: Open a new conversation, select Goal Mode from the input field, set a suggested or custom goal, add inputs and completion criteria, and submit.
Tip: Write the goal not just as a task, but with acceptance evidence such as a file location, test result, or source link.
✓ Verified by official sourceSource: How to Use AutoClaw Goal Mode for Multi-Step TasksAdding a network secret for the same host no longer overwrites the existing secret
Adding a new secret to a host that already has a network secret no longer silently replaces the existing value; the Add secret form rejects it and shows the existing secret.
Usage details
Who can use it: Organization administrators managing network secrets in cloud environments.
How to use it: In the Add secret form, select an existing host and try submitting a new secret; follow the form’s rejection and the link to the existing secret.
Tip: When rotating a secret, use the management workflow that deliberately updates the existing secret instead of adding a new entry.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesCorrupted configuration files are no longer overwritten
ZCode no longer automatically overwrites corrupted configuration files. This change reduces the risk of configuration-related data loss.
Usage details
Who can use it: All users of the ZCode 3.17.2 desktop version.
How to use it: Update ZCode to 3.17.2; if you see a corrupted configuration warning, save a copy of the file first instead of deleting or overwriting it.
Tip: Keep backups of custom settings and MCP configurations in a separate folder before upgrading.
✓ Verified by official sourceSource: ZCode Sürümler ve Güncellemeler — v3.17.2SendMessage reaches a running background agent sooner
SendMessage sent to a running background agent now moves it to the background without stopping its command, allowing the message to arrive sooner.
Usage details
Who can use it: Claude Code agent team workflows that use background agents and SendMessage.
How to use it: Send SendMessage while a background agent is running; watch to confirm the running command is moved to the background without stopping and the agent receives the message.
Tip: Use this when you don’t want the command interrupted; choose the separate Stop behavior for an urgent stop.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesOfficial remote CMS MCP connection for Cloud Monitor released
Alibaba Cloud CMS MCP exposes Cloud Monitor operations as remote MCP tools; alert, metric, trace, event, entity, Prometheus, and Grafana surfaces can be discovered in the default `core` ToolSet. The same operation contract as the CLI is preserved, without requiring `aliyun cms2` to be installed on the client machine.
Usage details
Who can use it: Alibaba Cloud CMS users; teams using Cursor, Claude Code, Lingma, or their own MCP client. China and international endpoints are available; write operations require appropriate RAM permissions.
How to use it: Add the region-appropriate Streamable HTTP `/mcp` endpoint to your MCP client with `?toolsets=core`; use `core,notification-channel` if needed. Authenticate with OAuth or an existing Alibaba Cloud session, then run a read-only workspace or alert query first.
Tip: Don’t put an AccessKey in mcp.json; connect with a read-only RAM role first, then enable the required write ToolSets.
✓ Verified by official sourceSource: Friday, 21:47, Checkout Is Slow Again: Twenty Minutes of Troubleshooting with CMS MCPMCP connection requests appear as cards in cloud sessions
Cloud sessions now display a visible card for a link that an MCP server requests to open; users can open or reject the link.
Usage details
Who can use it: Claude Code cloud sessions where MCP servers request to open a link.
How to use it: Run the MCP server in a cloud session; use the visible link card to open the URL or reject the request.
Tip: Don’t open the link before verifying the domain on the card; make sure it belongs to the expected MCP server integration.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesMulti-session cursor support for Computer Use
Computer Use now supports separate cursors for multiple sessions. This reduces interference between simultaneous or parallel computer interactions.
Usage details
Who can use it: ZCode v3.17.2 and later; users who use Computer Use.
How to use it: Start multiple Computer Use sessions and carry out tasks using each session’s own cursor.
Tip: Assign different tasks to each session and clearly specify its target application.
✓ Verified by official sourceSource: ZCode Releases & Updates — v3.17.2Change Request and Git worktree workflows are integrated with code review
The experimental Change Request & Worktree feature lets you review local and PR diffs using a searchable file tree and file-by-file navigation. Selected code and comments can be sent to the Agent, and checks and review status can be tracked.
Usage details
Who can use it: DeepSeek Harness v0.2.2-alpha.3; experimental feature for local Git repositories and PR workflows.
How to use it: Update to the latest version, open the Change Request & Worktree entry, and select a local or PR diff from the file tree; send selected code/comments to the Agent and track the checks.
Tip: Keep the worktree separate before reviewing, and reduce context costs by sending the Agent only relevant files and comments.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-alpha.3 sürüm notları`claude -p` now exits with code 1 when disk space or quota runs out
If a transcript or `~/.claude.json` can’t be written because the disk is full or a quota has been reached, `claude -p` no longer returns success code 0; it prints an error and returns exit code 1 in text and JSON modes.
Usage details
Who can use it: All users running automation with `claude -p`; some full NFS exports may be excluded.
How to use it: Check the exit code of the `claude -p` call and log the error message when it returns 1; configure your CI workflow to fail on this code.
Tip: In automation, rely on the exit code as well as stdout, and monitor disk space and quotas in the transcript directory.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesRead-before-write and external-change checks are disabled by default for file changes
Read-before-write and external-change protections are disabled by default; sandbox boundaries and the Review workflow remain in place. Profiles that want the old protections can explicitly restore the observation policy.
Usage details
Who can use it: DeepSeek Harness v0.2.2-alpha.3; profile-based file safety behavior.
How to use it: Open your profile’s observation policy and explicitly enable the previous file protections; if you use the default behavior, approve changes in the Review view.
Tip: Restore the previous read-before-write and external-change checks in production or shared repositories; don’t skip the Review check when prototyping quickly.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-alpha.3 sürüm notlarıConsent added for tool calls that send email addresses over the network
If a web or shell tool is going to send the account’s email address over the network, Claude Code now asks for consent before making the call.
Usage details
Who can use it: Sessions using Claude Code’s web or shell tools.
How to use it: Run the web or shell tool; if the email address will be included in a network request, review the permission prompt and approve or deny it.
Tip: To prevent unnecessary identity sharing, deny requests to unexpected domains.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesDate-range cost reporting added to the gateway admin API
The Claude apps gateway admin API now returns historical spending by total, developer, day, or IdP group through `GET /v1/organizations/analytics/cost`.
Usage details
Who can use it: Claude apps gateway administrators; organizations with access to the admin API.
How to use it: Call `GET /v1/organizations/analytics/cost` from your authorized gateway admin client and request a date range and breakdown by total, developer, day, or IdP group.
Tip: Export cost reports regularly with IdP group and daily breakdowns to catch budget deviations early.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesGo SDK expands GPT-Live Responses delegation tools
OpenAI Go SDK 3.92.0 adds a new API surface for GPT-Live’s Responses-based delegation tools. This release makes it easier to use Go for the delegation flow that lets GPT-Live sessions work with Responses-side hosted tools, MCP, and the application’s own tools.
Usage details
Who can use it: OpenAI Go SDK users; v3.92.0 is available for Go applications using GPT-Live/Responses delegation.
How to use it: Run `go get github.com/openai/openai-go/v3@v3.92.0`; use the `delegation.responses.tools` configuration in a GPT-Live session and send the required tool results back with the matching `call_id`.
Tip: Keep delegation separate from client-side tool calls; for long-running tasks, explicitly monitor the backend model, tool approvals, and error events.
✓ Verified by official sourceSource: OpenAI Go SDK v3.92.0 sürüm notlarıSource: GPT-Live delegation ve araçları kılavuzuNested `claude -p` calls no longer corrupt the outer transcript
Fixed an issue where a `claude -p` call running in a Cloud or self-hosted runner session wrote to the transcript using the outer session ID and resumed the wrong conversation on restart.
Usage details
Who can use it: Automations running `claude -p` inside a Cloud or self-hosted runner.
How to use it: Run nested `claude -p` in a runner session, then restart the outer session and check that the transcripts are separate.
Tip: Track inner and outer workflows using separate IDs; if you need a shared transcript, configure this explicitly.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesGemini products will be off by default for K-12 student accounts until an administrator enables them
Google announced that access to Gemini, Gemini Notebook, Gemini in Classroom, and Google Search will be off by default for K-12 student accounts. These products will not be available to student accounts unless a school administrator explicitly enables them; rollout details will be shared in the coming months.
Usage details
Who can use it: K-12 student accounts and school administrators; the effective date has not yet been finalized.
How to use it: When the rollout begins, school administrators must explicitly enable each AI product in the admin controls; Google has not yet published the exact menu path or implementation date.
Tip: Schools should prepare product-specific permission policies and inform teachers now so classroom workflows are not disrupted during the transition to the default-off model.
✓ Verified by official sourceSource: Google: Öğrenme için sorumlu AI tasarımıMCP settings detection restricted to registered workspaces
Access to MCP settings detection has been restricted to registered workspaces only. This reduces the chance of settings in other locations being read accidentally.
Usage details
Who can use it: ZCode v3.17.2 and later; workspaces where MCP settings are detected.
How to use it: Register the workspace in ZCode, then have it detect MCP settings through the configuration interface.
Tip: Keep MCP configuration files only in the workspaces where you need them.
✓ Verified by official sourceSource: ZCode Releases & Updates — v3.17.2Daily bill and discount breakdown query added to MiMo API
The new GET API returns the daily bill by currency as the original amount, discount, gift credit deduction, and final amount paid. It covers API call usage and gift credit deductions, making financial reconciliation easier.
Usage details
Who can use it: Xiaomi MiMo API accounts with a standard `sk-` API Key; according to the official documentation, the earliest queryable date is 2026-09-01.
How to use it: Open MiMo API Docs > API Reference > Finance > GET Query Daily Bill; use UTC dates and the `Authorization: Bearer $MIMO_API_KEY` header in the `GET https://api.xiaomimimo.com/v1/platform/billing/daily?start_date=YYYY-MM-DD&end_date=YYYY-MM-DD` call. Keep the query range to a maximum of 31 days.
Tip: For reconciliation, use the formula `final_paid_amount = original_amount - discount_amount - gift_amount`; match token/request records from the usage API with this bill response by date and currency.
✓ Verified by official sourceSource: MiMo API günlük fatura sorgusu dokümanıSource: MiMo API Platform V2.2.0 günlüğüMiMo Code removes cloud session sharing and third-party uploads
MiMo Code no longer offers cloud session sharing, automatic synchronization, unsharing, or sharing URL import; related fields and operations have been removed from the CLI, TUI, HTTP, GitHub integration, plugin API, and SDKs. Legacy configuration and local databases remain usable, but local sharing records are cleared, and previously published third-party links are not revoked.
Usage details
Who can use it: Both MiMo Code editions and all release channels; local JSON export/import remains available, while Desktop LAN sharing is outside the scope of this commit.
How to use it: There is no way to enable it: do not use `/share`, `/unshare`, or sharing URL import; these paths have been removed. Use local JSON export/import to move sessions.
Tip: Export sessions as local JSON before migrating rather than relying on old cloud links; separately check the status of published third-party links.
✓ Verified by official sourceSource: MiMo Code GitHub commit: remove cloud session sharing and third-party uploadsStart MCP servers on computers without Node installed
MCP servers can now start normally on computers without Node installed. This reduces setup barriers on remote or restricted machines that use MCP.
Usage details
Who can use it: ZCode v3.17.2 and later; computers that use MCP servers. No separate operating system or plan restriction was specified.
How to use it: Add the MCP server to your ZCode configuration and start it; Node does not need to be installed separately on the target computer.
Tip: Still check the MCP server’s own runtime requirements in its installation notes.
✓ Verified by official sourceSource: ZCode Releases & Updates — v3.17.2Off-peak packages no longer renew automatically when they expire
Off-peak packages no longer renew automatically after they expire. This reduces the risk of unwanted renewal charges for expired packages.
Usage details
Who can use it: ZCode Off-peak package users; v3.17.2 and later. Regional payment terms were not specified.
How to use it: Use the package as usual; when it expires, check the plan or payment screen to confirm that it has not renewed automatically.
Tip: Still check your account and payment history regularly to verify the package status.
✓ Verified by official sourceSource: ZCode Releases & Updates — v3.17.2Session ownership corrected and content leak risk addressed
Fixed a bug that caused ownership mix-ups when switching between or restoring sessions. This prevents the contents of one session from leaking into another.
Usage details
Who can use it: All users of the ZCode 3.17.2 desktop version.
How to use it: Update ZCode to 3.17.2; after opening and closing multiple sessions, check each session’s tasks and content separately.
Tip: On shared devices, review old sessions after updating and close any sessions you no longer need.
✓ Verified by official sourceSource: ZCode Sürümler ve Güncellemeler — v3.17.2Permission confirmation required for scripts in custom commands
Scripts run by custom commands now require permission confirmation before execution. This reduces the risk of unexpected code execution through custom commands.
Usage details
Who can use it: ZCode v3.17.2 and later; users who use custom commands.
How to use it: Run the custom command, review the permission prompt shown, and approve only the expected script.
Tip: Regularly review the script contents and file access of custom commands you use frequently.
✓ Verified by official sourceSource: ZCode Releases & Updates — v3.17.2High-risk actions can no longer be approved in Plan Mode
The behavior that allowed high-risk actions to be approved accidentally in Plan Mode has been fixed. These actions are now blocked under plan constraints.
Usage details
Who can use it: ZCode v3.17.2 and later; all users who use Plan Mode.
How to use it: Run a task in Plan Mode and confirm that the approval flow blocks the action when a high-risk action is proposed.
Tip: Use the planning phase only for review and preparation; initiate actual changes in a separate, controlled execution phase.
✓ Verified by official sourceSource: ZCode Releases & Updates — v3.17.2Self-hosted sessions restarted during plan approval retain recent turns
Fixed an issue where self-hosted runner sessions restarted while awaiting plan approval could lose recent turns or the entire conversation.
Usage details
Who can use it: Claude Code sessions with plan approval on self-hosted runner.
How to use it: Restart the session during plan approval and check that the latest messages and approval status are retained in the transcript.
Tip: Still, back up the transcript before restarting during critical plan approvals.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesPython SDK expands GPT-Live Responses delegation tools
OpenAI Python SDK 3.43.0 expands tool support in GPT-Live’s Responses-based delegation flow. The official GPT-Live guide explains that options such as hosted tools, MCP, tool search, programmatic tool calling, custom functions, computer use, local shell, and apply patch can be configured under delegation.responses.tools.
Usage details
Who can use it: OpenAI Python SDK users; v3.43.0 is available for applications using GPT-Live/Responses delegation.
How to use it: Upgrade the SDK with `pip install -U openai==3.43.0`; add the required tools to `delegation.responses.tools` in a GPT-Live session, then set `tool_choice` and `parallel_tool_calls` as needed.
Tip: Handle tool calls together with your application’s approval and result-return flow; in particular, provide MCP credentials again for update requests.
✓ Verified by official sourceSource: OpenAI Python SDK v3.43.0 sürüm notlarıSource: GPT-Live delegation ve araçları kılavuzuQoder Fast mode offers a one-click fast task experience
Qoder Fast was introduced as a new tier that delivers a fast task experience without asking users to select model and reasoning settings individually. In a 10-task test shared by Alibaba, Fast showed a 152% increase in generation speed over the previous Performance tier and took about half as long per task as Auto; quality scores were similar.
Usage details
Who can use it: Across the Qoder product family; access may vary by product, account, and current model configuration.
How to use it: Open the model or working tier selector in the Qoder product you use, select Fast, and start the task.
Tip: Choose Fast for quick idea validation, local code changes, and short feedback loops; use a stronger tier for deep architectural work.
✓ Verified by official sourceSource: Qoder Fast modu duyurusuQoder Lite launches worldwide for $10 a month with 1,000 credits
Qoder Lite began offering 1,000 monthly credits and access to selected cost-effective models for $10 a month worldwide. The plan includes Repo Wiki, Cloud Agents, and other Qoder features; a promotion offering 1,000 bonus credits on the first purchase and 500 on renewal was also announced.
Usage details
Who can use it: Individual users with a Qoder Global account worldwide; Lite, Pro, and higher-tier plans are available.
How to use it: Open https://qoder.com/pricing with a Qoder Global account, click the Subscribe button for the Lite plan, and complete checkout.
Tip: Preserve credits by choosing a cost-effective model such as Qwen3.8-Flash for routine tasks and a flagship model for long, complex tasks.
✓ Verified by official sourceModels: Qwen3.8-Flash, Qwen3.8-Max, DeepSeek-V4.1-Flash-1017, GLM-5.3-Flash, DeepSeek-V4-Pro, GLM-5.3, Kimi-K3, Kimi-K2.8-Preview, MiniMax-M3Source: Qoder Lite plan duyurusuSource: Qoder güncel fiyatlandırmasıCommand, plugin, and approval capabilities added to the TypeScript and Python SDKs
The TypeScript and Python SDKs now support native commands and plugin management. Plan review, continuing toward a goal, and action approval are supported; the SDK must be restarted for plugin changes to take effect.
Usage details
Who can use it: TypeScript and Python SDKs included in DeepSeek Harness v0.2.2-alpha.3.
How to use it: Update the SDK packages to versions compatible with v0.2.2-alpha.3, use native command/plugin management and approval workflows, and restart the SDK process after changing a plugin.
Tip: Require plan review and action approval for automated tasks; deploy plugin updates as a separate restart step.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-alpha.3 sürüm notlarıThe disabled-by-default apply_patch tool supports multi-file changes
The new `apply_patch` tool can create, edit, delete, or move multiple files in a single call. You can review changes in the Review view, and the tool is disabled by default.
Usage details
Who can use it: DeepSeek Harness v0.2.2-alpha.3; optional tool enabled per profile.
How to use it: Update to the latest version, explicitly enable the `apply_patch` tool in your profile, then request a multi-file change and open the Review view before applying it.
Tip: Enable the tool only in trusted repositories, and pay particular attention to file deletions and moves before accepting the unified patch.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-alpha.3 sürüm notlarıAccess scope for local web previews narrowed
The scope of local web page previews’ access has been restricted for security. This change reduces unnecessary access to local files and resources.
Usage details
Who can use it: ZCode v3.17.2 and later; users who use local web previews.
How to use it: Open the local web preview as usual, then verify that it accesses only resources within the permitted scope.
Tip: If the preview does not work, move the required resource into the saved workspace instead of sharing files broadly.
✓ Verified by official sourceSource: ZCode Releases & Updates — v3.17.2Scheduled tasks moved to unified Host management
Scheduled tasks are now managed through a single Host-level workflow. Users can create a task manually, edit it directly, or create it from a prefilled prompt in a session; drafts and conflict review are preserved.
Usage details
Who can use it: DeepSeek Harness v0.2.2-alpha.3; scheduled tasks run in reusable or new sessions.
How to use it: Open Automation/Scheduled tasks to create or edit a task; optionally create one from a prefilled prompt in a session, then run it with an existing or new session.
Tip: For complex scheduling expressions, save the draft and complete the conflict review; for simple name changes, keep the existing scheduling rule.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-alpha.3 sürüm notlarıSupport for macOS 12 and earlier ends with ZCode 3.17.2
ZCode 3.17.2 and later macOS releases are based on Electron 44 and Chromium 152; the minimum operating system is macOS 13 Ventura. macOS 12 (Monterey) and earlier are no longer supported.
Usage details
Who can use it: ZCode 3.17.2 and later macOS releases; macOS 13 Ventura and later are supported, while macOS 12 and earlier are not.
How to use it: Check your macOS version under Apple menu > About This Mac; if necessary, upgrade to Ventura or later, then download ZCode 3.17.2 from the official site.
Tip: Back up your project folders and ZCode settings before upgrading the operating system.
✓ Verified by official sourceSource: ZCode Sürümler ve Güncellemeler — v3.17.2ZCode automatically resumes workflows paused by quota limits
Workflows paused because of quota limits now resume automatically when the quota resets. Users do not need to restart the task manually.
Usage details
Who can use it: ZCode v3.17.2 and later; workflows that use quotas. Quota terms may vary by plan.
How to use it: Run a workflow as usual. If it pauses because of a quota limit, expect the same workflow to resume automatically when the quota renews.
Tip: For long workflows, monitor quota renewal times and any additional usage caused by automatic resumption.
✓ Verified by official sourceSource: ZCode Releases & Updates — v3.17.2One-click website deployment with the ZCode Sites plugin
ZCode v3.17.2 adds the official ZCode Sites plugin. The plugin supports deploying generated websites with one click.
Usage details
Who can use it: Desktop users running ZCode v3.17.2; no plan or platform restrictions specified.
How to use it: Install the official ZCode Sites plugin from the plugin store, then use the plugin’s one-click website deployment action.
Tip: Before deployment, check the preview and the files that will be sent to the production environment.
✓ Verified by official sourceSource: ZCode Releases & Updates — v3.17.2Kimi K3.1 Pro and MiniMax M3.1 Flash settings in ZCode
New model configuration options have been added to ZCode for Kimi K3.1 Pro, MiniMax M3.1 Flash, and other models. Users can now select these providers and models in the configuration flow.
Usage details
Who can use it: ZCode v3.17.2 and later; users with access to supported models. No separate plan or country list was provided.
How to use it: Open the model configuration screen or model picker, enter the required access details for Kimi K3.1 Pro or MiniMax M3.1 Flash, and save.
Tip: For long coding tasks, compare context windows, speed, and quota limits before selecting a model.
✓ Verified by official sourceModels: Kimi K3.1 Pro, MiniMax M3.1 FlashSource: ZCode Releases & Updates — v3.17.2`/design-sync` gives an early warning when a design system has moved
`/design-sync` now tells you upfront that the design system has moved to Claude and stops before starting production instead of failing during upload.
Usage details
Who can use it: Claude Code and design system workflows that use `/design-sync`.
How to use it: Run `/design-sync`; if the design system has moved, read the notification and configure its new location in Claude first.
Tip: Before a large sync, check the design system source’s active location to avoid unnecessary uploads.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesWon’t delete the wrong guide when no `/team-onboarding` name is specified
Fixed an issue where Claude deleted the most recently updated guide if no guide was specified when multiple guides existed.
Usage details
Who can use it: Claude Code teams using `/team-onboarding` guides.
How to use it: Explicitly provide the guide name before deleting; with v2.1.317, verify that an unnamed deletion request doesn’t target the wrong guide.
Tip: For bulk guide management, list guides first and verify their names explicitly before deleting.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notes`claude auth status` no longer drops the session during a slow token refresh
`claude auth status` no longer exits before a slow token refresh is recorded, so the user no longer appears to be accidentally signed out on the next run.
Usage details
Who can use it: All platforms using Claude Code authentication.
How to use it: Update Claude Code to v2.1.317 and run the `claude auth status` command.
Tip: In CI or managed environments, don’t terminate the process immediately after checking authentication status; allow the token refresh to complete.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesBackground session permission mode no longer gets lost during promotion
Fixed an issue where a recent permission mode change was lost when a background session was promoted or retired.
Usage details
Who can use it: Users changing permission modes in background Claude Code sessions.
How to use it: Change the background session’s permission mode, then promote or retire the session; check that the new mode is retained.
Tip: Keep changes to authorized modes auditable by recording them in the transcript or event log.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesAuto mode makes a nonresponsive subagent check visible
If the Auto mode check cannot respond to a subagent’s tool call, Claude Code now displays a notification in the main view. Previously, the user might not have seen anything in this situation.
Usage details
Who can use it: Claude Code sessions using Auto mode and subagents.
How to use it: In Auto mode, run a subagent that makes a tool call; if the check does not respond, follow the notification in the main view.
Tip: When you see the notification, check that the subagent isn’t stuck waiting and manually restart the call if needed.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesAuto mode no longer loops on gateway 403 and 422 security responses
Fixed an issue where Auto mode failed on every turn when an LLM gateway used as the base URL returned 403 or 422 for a server-side security check. The behavior for 403 responses returned by `apiKeyHelper` is unchanged.
Usage details
Who can use it: Auto mode sessions using an LLM gateway or custom base URL.
How to use it: Configure the gateway’s base URL and run Auto mode; check the new error behavior and gateway logs for 403/422 responses.
Tip: Verify the expected status codes and authentication path for the gateway’s security check in advance.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesThe standalone Git worktree extension has moved to the unified workflow
The Git worktree entry is now part of the Change Request & Worktree workflow. Users who previously enabled the standalone extension need to remove the old `@deepseek-ai/dsh-experimental-tool-worktree` entry from their profile.
Usage details
Who can use it: DeepSeek Harness v0.2.2-alpha.3; profiles using the old standalone worktree extension.
How to use it: Open your profile file and remove the `@deepseek-ai/dsh-experimental-tool-worktree` entry from `dsh.profile.bundles`; then use the unified Change Request & Worktree entry.
Tip: Back up your profile before migrating; don’t keep the old and unified worktree entries enabled at the same time.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-alpha.3 sürüm notlarıThe Bedrock ‘Input is too long’ error is no longer hidden
The gateway no longer converts Amazon Bedrock’s `Input is too long.` error into a generic upstream error, making it possible to understand why the session needs automatic compaction.
Usage details
Who can use it: Organizations using Amazon Bedrock through the Claude apps gateway.
How to use it: When an error occurs in a long-context Bedrock session, inspect the `Input is too long.` message in the gateway response and the compaction behavior.
Tip: If you see the error, reduce the context or review your compaction settings; don’t retry as if it were a generic upstream error.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesDocument viewing is isolated by session, and plugin processes are shared
The document viewer now runs in isolation for each session. Multiple document plugins also share the same process to reduce resource consumption.
Usage details
Who can use it: Users of the ZCode 3.17.2 desktop version.
How to use it: Upgrade ZCode to 3.17.2; open documents in different sessions to use each session’s viewing state separately.
Tip: Closing unnecessary sessions while viewing many documents increases the resource savings from process sharing.
✓ Verified by official sourceSource: ZCode Sürümler ve Güncellemeler — v3.17.2Region or network blocks are no longer mistaken for sign-in errors
HTTP 403 errors caused by a region or network block are no longer shown as sign-in failures; the error message now explains the block itself.
Usage details
Who can use it: Users accessing Claude Code with region restrictions or network filtering.
How to use it: Update Claude Code and retry the connection; use the region or network block details in the new error message to resolve the issue with your administrator.
Tip: Check allowed domains and region policies on your corporate proxy before changing credentials.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesAn empty Bedrock bearer token setting no longer overrides machine credentials
When the environment configuration for a Stdio MCP server leaves `AWS_BEARER_TOKEN_BEDROCK` empty, the variable is no longer passed through as empty; the server can use the machine’s AWS credentials.
Usage details
Who can use it: Stdio MCP servers using Bedrock.
How to use it: Leave `AWS_BEARER_TOKEN_BEDROCK` empty in the MCP environment configuration and test with v2.1.317 that the server connects using the machine’s identity.
Tip: Document the credential source explicitly; monitor logs for unexpected switching between an empty variable and a real bearer token.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesPrompt overflow after compaction is fixed when a large CLAUDE.md changes
The `Prompt is too long` error after automatic compaction has been fixed for cases where a very large `CLAUDE.md` file changes after the last submission.
Usage details
Who can use it: Long Claude Code sessions using a large CLAUDE.md.
How to use it: Update the CLAUDE.md file, trigger compaction in a long session, and retry the new turn.
Tip: Avoid bloating the root instruction file with unnecessary repetition; move long rules into a skill or reference file.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesSkills synced with claude.ai no longer disappear when writing to disk fails
When skills synced from claude.ai can’t be written to disk, they no longer disappear and enter a re-download loop every ten minutes.
Usage details
Who can use it: Users who sync claude.ai skills with Claude Code.
How to use it: Leave syncing enabled, update Claude Code to v2.1.317, and verify that the skill remains listed after restarting the session.
Tip: Fix disk permissions too; the persistence fix doesn’t store the skill locally when the environment lacks write permissions.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesCloud Auto mode no longer requests unnecessary confirmation for Artifact output operations
In cloud sessions, file uploads, database writes, and saves to the outputs folder by the Artifact tool no longer unnecessarily pause Auto mode for confirmation.
Usage details
Who can use it: Auto mode and the Artifact tool in Claude Code cloud sessions.
How to use it: Enable Auto mode in a cloud session, then use Artifact to upload an output, write to a database, or save to the outputs folder.
Tip: Still review your permission policy before running Artifact operations that write data to external systems with production data.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesCode tab shows when remote managed settings are loaded
The Claude Desktop Code tab warns if the organization’s remote managed settings have not yet loaded for the session and shows a second notification when they have loaded.
Usage details
Who can use it: Claude Desktop Code sessions configured with organization-managed remote settings.
How to use it: Start a new session in the Code tab in Claude Desktop; wait for the settings-loaded notification and verify that organizational policies have been applied.
Tip: Don’t start policy-governed tasks before the initial notification clears; wait for the notification confirming that the settings have loaded.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesMemory usage and startup improved in headless sessions with many plugins
High memory usage and slow startup were fixed in headless sessions using many plugins that register commands.
Usage details
Who can use it: Headless Claude Code and automations using many plugins.
How to use it: Start a headless session with plugins enabled using v2.1.317, and monitor startup time and memory usage.
Tip: Reduce startup overhead further by enabling only the plugins needed for headless deployments.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesPrevious summaries are preserved during compaction in very long conversations
Fixed an issue that caused very long conversations to lose their previous summary when compaction had to shorten its own request.
Usage details
Who can use it: Long-context Claude Code conversations.
How to use it: Trigger compaction in a long conversation and check that the previous work summary remains in the new context.
Tip: Also write critical decisions in a short project note; even with improved compaction, don’t rely on a single context store.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesLocal and project MCP servers no longer disconnect in Desktop sessions
Fixed connection behavior for project and local MCP servers that disconnected after the first turn in some Claude Desktop sessions.
Usage details
Who can use it: Users of project or local MCP servers in Claude Desktop.
How to use it: Update Claude Desktop to v2.1.317 and reconnect your MCP server; check the tool list in subsequent turns.
Tip: After updating, try calling tools over multiple turns to verify that the MCP connection really persists.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesDisabled scheduled tasks can be listed and deleted
Fixed an issue where stored tasks could not be listed or deleted after the scheduled task feature was turned off.
Usage details
Who can use it: Claude Code cloud scheduled task administrators.
How to use it: Temporarily turn off scheduled tasks, open the task list, and list or delete a stored task.
Tip: Export tasks before turning off the feature; verify in the new flow that you can still clean them up while it is disabled.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesFile previews now include a persistent folder tree and file location
File previews now display a persistent folder tree and let you jump to the open file’s location in the tree. This makes it easier to navigate files in the same folder.
Usage details
Who can use it: DeepSeek Harness v0.2.2-alpha.3 preview release.
How to use it: Open a text or Markdown file in the preview, use the pinned folder tree, and run the action to find/locate the current file in the tree.
Tip: In large repositories, open the folder tree first and locate the target file; keep the preview context rather than searching for files from chat.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-alpha.3 sürüm notlarıOptions to reveal files in Finder/Explorer
File cards, inline links, and the open menu now offer an option to reveal a file in Finder or Explorer.
Usage details
Who can use it: ZCode v3.17.2 and later; users of the macOS and Windows file explorers.
How to use it: Right-click the file in a file card, inline link, or the open menu, then select Show in Finder/Explorer.
Tip: If there are multiple similar files, reveal the file’s location to quickly verify the correct output.
✓ Verified by official sourceSource: ZCode Releases & Updates — v3.17.2GPT-6.1 Terra joins the Dots model family
OpenAI added GPT-6.1 Terra to the frontier model family powering Dots in an update to its Dots announcement on November 3. According to the announcement, a dot can use different models for different parts of its work; this is a new detail that expands the model coverage available to existing Dots users.
Usage details
Who can use it: Pro and Business Premium users with Dots access in eligible markets; administrator-enabled beta terms remain in place for Enterprise, Edu, and Healthcare.
How to use it: Create a dot or open an existing one in the ChatGPT desktop app or on desktop web; no separate model-selection step is specified for GPT-6.1 Terra, which is used automatically once the support rollout reaches your account.
Tip: Give the dot visual tasks, computer use, and general planning together; review critical side effects in Activity View and don’t leave them unapproved.
✓ Verified by official sourceModels: GPT-6.1 Terra, GPT-6 AstraSource: Dots duyurusu ve 3 Kasım güncellemesiDetailed diagnostics and the Host log are shown for plugin installation errors
Plugin installation errors now show full diagnostic details and the location of the Host log. Installation failures are distinguished from failures to enable a plugin after installation.
Usage details
Who can use it: DeepSeek Harness v0.2.2-alpha.3; plugin manager.
How to use it: When a plugin installation fails, open the error details, check the specified Host log, and determine whether the issue occurred during installation or activation.
Tip: In support requests, share the full diagnostic text and Host log path, not just the error summary.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-alpha.3 sürüm notlarıPlugins can provide dynamic pi-ai headers based on session and request purpose
Plugins can now send dynamic pi-ai HTTP headers based on the session ID and request purpose. Invalid headers are rejected before the request is sent.
Usage details
Who can use it: DeepSeek Harness v0.2.2-alpha.3; plugin developers using pi-ai.
How to use it: Update your plugin for the alpha.3 API; tie header generation to the Session ID and request purpose, and handle validation errors before sending the request.
Tip: Generate identity or authorization headers from session context instead of hardcoding them; catch invalid headers locally rather than leaving them to fail at the remote API.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-alpha.3 sürüm notlarıScreen reader no longer announces subagent counters on every change
Fixed an issue where a subagent running in screen reader mode repeatedly announced elapsed time and token counts after every change.
Usage details
Who can use it: Accessibility users who use Claude Code screen reader mode.
How to use it: Turn on screen reader mode and run a subagent; check that elapsed time and token information are not spoken repeatedly without need.
Tip: For long tasks, keep screen reader notifications low-noise so you hear only status changes.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesConnection floods to a blocked HTTP OTel proxy are prevented
Fixed repeated connection attempts when a proxy for an HTTP OpenTelemetry endpoint returns HTTP 403.
Usage details
Who can use it: Claude Code environments using an HTTP OTLP endpoint and a proxy that returns 403.
How to use it: Configure an OTel HTTP endpoint and reproduce the proxy block; confirm in the logs with v2.1.317 that the repeated connection flood has stopped.
Tip: Fix proxy permissions or the OTLP URL instead of retrying a blocked endpoint.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesInvisible text watermarking is being extended to older Claude models
According to a report by Mashable citing an email to Anthropic users, starting November 3, outputs from Claude Mythos 5, Claude Opus 4.7, Claude Opus 4.6, and Claude Sonnet 4.6 will also carry an invisible text watermark. Anthropic’s official help page confirms work to add marking to older models, but does not officially state that activation for these four models is dated November 3; this report should therefore be considered unverified.
Usage details
Who can use it: According to the report, globally in Claude outputs from the older models in question and across surfaces where Claude is available; the watermark is invisible to readers and does not carry user or organization identity.
How to use it: There is no user-side setting to enable; generate text with the relevant model. Audit access is currently limited to a private preview for regulators, law enforcement, media, fact-checking organizations, researchers, and similar eligible organizations.
Tip: Assume Claude output may be watermarked before putting it into an enterprise text workflow; follow the official content marking and detection documentation.
! UnverifiedModels: Claude Mythos 5, Claude Opus 4.7, Claude Opus 4.6, Claude Sonnet 4.6Source: Anthropic expands text watermarking to older Claude versions beginning November 3Source: How Claude marks AI-generated contentSource: Claude text watermark açıklamasıIssues with legacy sessions, search, and loading compressed history fixed
Alpha.3 fixes opening and searching sessions that contain legacy subagent definitions, continuing a chat after loading compressed history, and correctly naming and sorting empty sessions. These are usability and backward-compatibility improvements rather than new features.
Usage details
Who can use it: DeepSeek Harness v0.2.2-alpha.3; current and legacy sessions.
How to use it: Update to the new version, open and search a legacy session, then send a new message after compressed history to use the recovery behavior.
Tip: Export important sessions before updating; if the issue persists, include the Host log along with the session ID.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-alpha.3 sürüm notlarıThe second call path remains visible in concurrent permission requests
The second of two permission requests opened at the same time no longer appears in Claude web and mobile apps without showing the file path or command.
Usage details
Who can use it: Concurrent tool permissions in Claude web and mobile apps.
How to use it: Trigger two tool calls at the same time; review the file path or command in the second permission card before deciding.
Tip: Don’t approve requests in bulk without seeing the permission card’s context; check each call’s path separately.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesSubagents using a different model now use their own effort level
Subagents and teammates using a different model from the main conversation now use their own saved level or the model default, instead of forcibly inheriting the session’s effort level.
Usage details
Who can use it: Claude Code users running subagents or teammates with multiple models.
How to use it: Assign the subagent a different model from the main model; check in the session view that it uses that model’s saved effort setting or default.
Tip: Test each model’s effort level separately to meet model-specific cost and quality goals.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesDaily IdP group records for developers are retained in gateways
Claude apps gateways with `admin:` enabled record developers’ IdP groups daily. Records are retained for `admin.identity_retention_days`, enabling group breakdowns in cost reports.
Usage details
Who can use it: Claude apps gateway organizations with the `admin:` setting enabled.
How to use it: Set `admin.identity_retention_days` in the gateway admin settings; use the IdP group breakdown in cost reports.
Tip: Set the identity data retention period to the minimum duration required by your organization’s compliance policy.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesSalesforce federated data store reaches GA in Gemini Enterprise
Connecting a Salesforce data source to Gemini Enterprise through federation, without importing it, is now generally available. Assistant can search accounts, contacts, opportunities, cases, leads, and custom objects; generate grounded responses; and create, update, or delete records.
Usage details
Who can use it: Gemini Enterprise organizations using Salesforce; the release notes cover Standard, Plus, Pay-as-you-go, and Frontline edition documentation.
How to use it: In the Gemini Enterprise documentation, open the Connectors > Salesforce > Set up a data store flow; authorize Salesforce and complete the federated data store configuration.
Tip: Before enabling write operations, use read-only queries to verify Salesforce permissions and which object fields will be exposed through federation.
✓ Verified by official sourceSource: Gemini Enterprise sürüm notlarıSource: Salesforce veri deposu kurma rehberiGo SDK can collect MCP argument progress
OpenAI Go SDK 3.92.0 adds support for collecting progress information for MCP tool call arguments in Responses streams. This SDK support passes intermediate progress to a Go application while the MCP call arguments are being generated.
Usage details
Who can use it: OpenAI Go SDK v3.92.0 users; Go applications using MCP tools in the Responses API.
How to use it: Upgrade with `go get github.com/openai/openai-go/v3@v3.92.0`; read the stream items related to MCP argument progress in your Responses WebSocket event loop and display them in the interface.
Tip: Use partial arguments only as a progress indicator; wait for complete and validated arguments before running the call.
✓ Verified by official sourceSource: OpenAI Go SDK v3.92.0 sürüm notlarıGo SDK adds support for typed existing-item references in Realtime
OpenAI Go SDK 3.92.0 adds support for typed existing-item references to items referenced in Realtime events. This lets developers handle Realtime inputs using specific reference types that the SDK can parse, rather than generic, loosely typed ones.
Usage details
Who can use it: OpenAI Go SDK v3.92.0 users; Go applications working with the Realtime API.
How to use it: Upgrade with `go get github.com/openai/openai-go/v3@v3.92.0`; parse Realtime events and item references through the new typed SDK models, and add a safe fallback for unknown item types.
Tip: Don’t force all Realtime items into the same structure; preserve type distinctions to prevent the application from breaking when new event types arrive.
✓ Verified by official sourceSource: OpenAI Go SDK v3.92.0 sürüm notlarıSource: Realtime API başlangıç kılavuzuGo SDK adds optional final response reconstruction for WebSocket
OpenAI Go SDK 3.92.0 adds optional final response reconstruction for the Responses WebSocket stream. Instead of merging individual delta events itself, an application can ask the SDK to reconstruct the completed response.
Usage details
Who can use it: OpenAI Go SDK v3.92.0 users; Go applications using Responses WebSocket.
How to use it: Upgrade with `go get github.com/openai/openai-go/v3@v3.92.0`; use the version’s opt-in final response reconstruction API with a Responses WebSocket connection and read the completed response with `FinalResponse(ctx)`.
Tip: Use raw delta events for low-latency interim output, and the reconstructed final response for validation and persistent storage.
✓ Verified by official sourceSource: OpenAI Go SDK v3.92.0 sürüm notlarıSource: Responses WebSocket modu kılavuzuCustom pre-meeting policy notice for Gemini in Google Meet
Administrators can set a short policy message and optional URL to display on the Green Room screen before participants join a meeting. They can choose to show the notice in every meeting or only when recording, transcription, or Gemini notes are enabled.
Usage details
Who can use it: Business Standard and Plus; Enterprise Standard and Plus; Education Fundamentals, Standard, and Plus; Frontline Plus; Google AI Pro for Education. The expanded rollout began on 2026-11-02.
How to use it: In the Admin console, enable the feature under Meet Safety Settings for the relevant OU; configure the Green Room message, URL, and display conditions.
Tip: Keep the policy message short and link to a detailed organization page; showing the notice only when Gemini notes or recording is enabled can reduce clutter on the join screen.
✓ Verified by official sourceSource: Google Workspace Updates: Meet öncesi özel politika bildirimiGuided Learning becomes globally available in Google Classroom
Guided Learning is now globally available in Google Classroom. A teacher-assigned activity works like a one-on-one study tool linked to class materials and gives teachers real-time insights into the topics where the class needs support.
Usage details
Who can use it: Global availability in Google Classroom; rollout and school administrator controls may vary by organization.
How to use it: A teacher assigns a Guided Learning activity in Google Classroom; the student opens the assigned activity to start the step-by-step study flow.
Tip: Link the activity to class materials; this helps Guided Learning guide students through the steps of thinking in the context of the lesson rather than simply giving them the answer.
✓ Verified by official sourceSource: Google: Öğrenme için sorumlu AI tasarımıGuided Learning begins piloting in Google Search for student accounts
Google is adding Guided Learning capabilities to Search for student accounts. When the system detects homework-like queries, it will guide students with questions and step-by-step prompts instead of jumping straight to the result; the pilot is planned to begin at select U.S. schools during the month.
Usage details
Who can use it: Select U.S. schools and student accounts during November 2026; not generally available.
How to use it: Submit a homework- or problem-type query using a participating school account; Search will show Guided Learning steps for eligible queries. There is no setting to enable it for general users yet.
Tip: Describe the question and what you are trying to understand in detail; complete each step of the guided flow rather than asking only for the final answer.
✓ Verified by official sourceSource: Google: Öğrenme için sorumlu AI tasarımıSecurity-blocked responses now return an error quickly instead of retrying
When a response is blocked by the security policy, ZCode now reports an error quickly instead of retrying repeatedly. This lets users see the reason for the block sooner.
Usage details
Who can use it: ZCode v3.17.2 and later; responses blocked by the security policy.
How to use it: Submit a task that may be blocked by the security policy; when a block occurs, review the error notification.
Tip: Instead of resubmitting the same request, narrow down the blocked content or action safely.
✓ Verified by official sourceSource: ZCode Releases & Updates — v3.17.2Repetitions without progress automatically stopped and reported
Repeated executions that make no progress are now automatically stopped, and users are notified. This is intended to limit wasted time and token usage.
Usage details
Who can use it: ZCode v3.17.2 and later; tasks experiencing repeated execution.
How to use it: Run the task as usual; if the system detects no progress, it will stop execution and show a notification.
Tip: After an automatic stop, narrow down the faulty step or target condition instead of repeating the task with the same instructions.
✓ Verified by official sourceSource: ZCode Releases & Updates — v3.17.2Ctrl+C clears the prompt line without stopping background agents
Fixed an issue that stopped all running background agents when Ctrl+C was pressed to clear text in the prompt line.
Usage details
Who can use it: Concurrent Claude Code sessions with background agents.
How to use it: While a background agent is running, type text in the prompt line and clear it with Ctrl+C; check that the agents continue running.
Tip: Use the explicit stop command to actually stop an agent; don’t mistake the prompt-clearing shortcut for a stop signal.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesInference within India is now available for Kimi K3 on Amazon Bedrock
According to AWS’s official announcement, Amazon Bedrock has made an in-country inference profile for Kimi K3 available. The `in.moonshotai.kimi-k3` profile processes requests in the Mumbai and Hyderabad regions, offering a new access option for workloads that require data residency in India.
Usage details
Who can use it: Amazon Bedrock customers; India geographic inference profile `in.moonshotai.kimi-k3`, with Mumbai and Hyderabad as target regions (AWS `ap-south-1` and `ap-south-2`). Kimi K3’s global and US profiles remain available as separate options.
How to use it: Enable access to Kimi K3 in Amazon Bedrock; use `in.moonshotai.kimi-k3` as the model ID in a Bedrock Runtime call, choose a source region such as `ap-south-1`, and grant the `bedrock:InvokeModel` and `bedrock:InvokeModelWithResponseStream` permissions in IAM.
Tip: If data must not leave India, use the India profile instead of `global.moonshotai.kimi-k3`; consider the global profile when broader access and cost are priorities.
✓ Verified by official sourceModels: Kimi K3Source: Introducing India-based inference for Kimi K3 on Amazon BedrockSource: Kimi K3 - Amazon Bedrock model cardTLS proxy connections work after root certificate renewal
Fixed the `unable to get local issuer certificate` API error that occurred after root certificate renewal with some TLS-inspecting proxies.
Usage details
Who can use it: Claude Code API clients behind TLS-inspecting proxies.
How to use it: Update Claude Code and make an API call behind the proxy; if needed, also renew the system root certificate store.
Tip: Test the proxy’s intermediate certificate chain and the client trust store together; don’t just change the API key.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesUser data migration now completes without an open project
User-level data migration can now complete even when no project is open. This removes the requirement to open a project during first launch or a version upgrade.
Usage details
Who can use it: Users upgrading to ZCode 3.17.2; no plan or operating system restrictions specified.
How to use it: Upgrade ZCode to 3.17.2 and launch the app without an open project; if a data migration prompt appears, complete the process.
Tip: Do not close the app during migration, and keep a backup of important user data.
✓ Verified by official sourceSource: ZCode Sürümler ve Güncellemeler — v3.17.2Viewing live metrics no longer starts recording automatically
In Meta VR Developer Hub 6.6.0, viewing live metrics no longer starts recording on its own. This prevents developers who only want real-time monitoring from generating unnecessary traces.
Usage details
Who can use it: Meta VR Developer Hub for Mac 6.6.0 and Performance Analyzer users.
How to use it: Open the live metrics view in Performance Analyzer; start recording separately when a trace is needed.
Tip: Use only live metrics to reduce storage consumption during short diagnostic sessions.
✓ Verified by official sourceSource: Meta VR Developer Hub 6.6.0 sürüm notlarımacOS Computer Use permissions are managed through a standalone app
macOS Computer Use now requests system permissions through a standalone app. The app opens the relevant system setting, provides drag-and-drop guidance, and refreshes permission status when you return.
Usage details
Who can use it: DeepSeek Harness v0.2.2-alpha.3; macOS Computer Use.
How to use it: Start Computer Use, go to the macOS system setting opened by the permission prompt, drag the indicated app into the relevant permissions list, and return to Harness.
Tip: Refresh the status after returning from the permissions screen; authorize the standalone app shown rather than running the app directly from the terminal.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-alpha.3 sürüm notlarıMarkdown file links support line ranges and column positions
Markdown file links in chat now accept line ranges and line-column suffixes. For invalid positions, the file still opens and the link remains a web link.
Usage details
Who can use it: DeepSeek Harness v0.2.2-alpha.3; chat file links.
How to use it: Use a file link in chat with a line range or line-column suffix; for example, click a Markdown file link that specifies the target location.
Tip: In code reviews, link only the relevant line range to give the Agent smaller, more focused context.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-alpha.3 sürüm notlarıMore detailed causes and solutions for MCP startup errors
MCP startup failures now show more specific causes and suggested solutions. Troubleshooting relies less on generic error messages.
Usage details
Who can use it: ZCode v3.17.2 and later; users who use MCP.
How to use it: Start an MCP server. If an error occurs, follow the displayed cause and solution steps to fix the MCP configuration.
Tip: Include the server name, workspace, and permission details from the error message in your support ticket.
✓ Verified by official sourceSource: ZCode Releases & Updates — v3.17.2UE 5.8 support added to Meta Haptics SDK for Unreal 207.0
Meta Haptics SDK for Unreal 207.0 adds support for a separate SDK package for Unreal Engine 5.8. The package provides the high-level API used to play and modify haptic clips created with Haptics Studio on Quest controllers.
Usage details
Who can use it: Developers using Unreal Engine for Quest; UE 5.8 projects.
How to use it: Download Meta Haptics SDK for Unreal version 207.0 from Meta VR Developers downloads and add it to your UE 5.8 project.
Tip: Match your project’s Unreal version to the SDK package’s UE 5.8 requirement; choose an earlier package version for older UE versions.
Meta Quest Developer Hub renamed Meta VR Developer Hub
With Meta Quest Developer Hub for Mac 6.6.0, the product is now called Meta VR Developer Hub (MVDH). The release refreshes the Casting Window design, moves Performance Analyzer trace capture to named presets, signs and notarizes macOS packages, and fixes issues with file management, installation, connectivity, and stability.
Usage details
Who can use it: Users developing Meta Quest apps on Mac; version 6.6.0.
How to use it: Download and install the Meta VR Developer Hub for Mac 6.6.0 package. Open Casting Window or Performance Analyzer to use the new interface and named trace capture presets.
Tip: Update automation and existing capture workflows to use the new named presets; check how the individual options changed in 6.6.0.
Microsoft Foundry credential error is now more informative
When Azure credentials can’t be loaded, the Microsoft Foundry error now provides refresh or troubleshooting steps and is reported as `cloud_credential_error`.
Usage details
Who can use it: Teams using Claude Code through Microsoft Foundry.
How to use it: Break Foundry authentication or try with an expired credential; refresh the credentials based on the `cloud_credential_error` code.
Tip: Classify this error code separately from sign-in errors in your monitoring system.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesOpenAPI for querying MiMo API account balance released
Xiaomi has added a new GET endpoint to programmatically read the availability and balance components of a standard API account. The response returns cash, gifts, frozen amount, credit limit, available credit, total available balance, and the account’s availability status.
Usage details
Who can use it: Xiaomi MiMo API accounts using a standard `sk-` API Key.
How to use it: Open MiMo API Docs > API Reference > Finance > GET Query Account Balance; in the `GET https://api.xiaomimimo.com/v1/platform/balance` call, send `Authorization: Bearer $MIMO_API_KEY` or the API key header shown in the documentation.
Tip: For balance alerts, evaluate `total_available` and `is_available` together rather than looking only at the cash balance; `total_available` also reflects credit and gift balances.
✓ Verified by official sourceSource: MiMo API hesap bakiyesi sorgusu dokümanıSource: MiMo API Platform V2.2.0 günlüğüOpenAPI for querying MiMo API Token Plan quota released
The remaining credits in a personal Token Plan package can now be read through the API. The response provides the plan code, period start and end, total/used/remaining credits, auto-renewal, and MiMo Claw status.
Usage details
Who can use it: Accounts with a personal Token Plan package that use a Token Plan-specific `tp-` API Key; the official documentation defines this endpoint for the personal edition package.
How to use it: Get the dedicated Base URL and Token Plan API Key from the Token Plan console; in the `GET https://token-plan-cn.xiaomimimo.com/v1/platform/token_plan` call, send `Authorization: Bearer $MIMO_API_KEY` or the API key header shown in the documentation.
Tip: For a usage bar and renewal alerts, monitor `remaining_credits` and `period.end_at` together; add a renewal reminder if `auto_renew` is off.
✓ Verified by official sourceSource: MiMo API Token Plan kredileri sorgusu dokümanıSource: MiMo API Platform V2.2.0 günlüğüDaily usage and token consumption query added to MiMo API
The new GET endpoint returns the day, request count, consumption amount, token breakdown, audio transcription duration, and inference type (`online`/`batch`) by model and add-on. Queries use UTC calendar days and cover a range of up to 31 days.
Usage details
Who can use it: Xiaomi MiMo API accounts with a standard `sk-` API Key; according to the official documentation, the earliest queryable date is 2026-09-01.
How to use it: Open MiMo API Docs > API Reference > Finance > GET Query Daily Consumption; use UTC dates and the `Authorization: Bearer $MIMO_API_KEY` header in the `GET https://api.xiaomimimo.com/v1/platform/usage/daily?start_date=YYYY-MM-DD&end_date=YYYY-MM-DD` call. Keep the interval between `end_date` and `start_date` to no more than 31 days.
Tip: For cost analysis, aggregate the `resource_type` and `inference_type` fields separately; this endpoint does not provide a breakdown of pre-discount amount, discount, and gift deductions, so also call the Daily Bill API for financial reconciliation.
✓ Verified by official sourceSource: MiMo API günlük kullanım sorgusu dokümanıSource: MiMo API Platform V2.2.0 günlüğüMiMo Code memory search supports Chinese substrings and short queries
Memory search now matches contiguous Chinese substrings, one- or two-character queries, and combinations of Han, Latin, and digits in both CLI and Desktop. Indexes for existing records are populated in the background in limited, resumable batches.
Usage details
Who can use it: MiMo Code CLI and Desktop; existing memory records are migrated automatically.
How to use it: Use the existing memory search function in MiMo Code without enabling a separate setting; enter your short Chinese query or substring directly.
Tip: For Chinese searches, using a contiguous, meaningful phrase produces more accurate results than scattered single-character matches.
✓ Verified by official sourceSource: MiMo Code GitHub commit: Chinese substring memory recallMiMo Code file tools preserve applicable diffs
Edit, Write, Apply Patch, and notebook-edit now emit applicable raw diffs; only the TUI and run output are truncated for display. This fixes issues where diffs could not be applied or reverted in indented blocks and prevents line-ending differences from being lost.
Usage details
Who can use it: The 2026-11-03 change on MiMo Code’s GitHub main branch; the latest official release, v0.1.16, is dated 2026-10-26, so no released version number is specified for this commit.
How to use it: Use the Edit, Write, Apply Patch, or notebook-edit tools in MiMo Code as usual; generated diffs can now be applied to and reverted from the target file.
Tip: For files using CRLF, apply the diff directly with the MiMo Code tool without reformatting it in a text editor.
✓ Verified by official sourceSource: MiMo Code GitHub commit: publish applicable file-tool diffsMiMo-based Super XiaoAi 2.0 accompanies the HyperOS 4 rollout
Xiaomi’s official account announced that the stable version of HyperOS 4 began rolling out via staged OTA updates to eligible devices on November 3. Android Central reported that Super XiaoAi 2.0 will run on Xiaomi MiMo in this release and support more complex tasks such as photo search, planning, smart home control, and on-screen product recognition; the official HyperOS page does not explicitly mention this MiMo connection.
Usage details
Who can use it: Global, phased rollout of HyperOS 4; the official device list includes early November, mid-November, and December waves. AI features may be limited to select devices, regions, and languages; the MiMo-based Super XiaoAi 2.0 detail is confirmed only by secondary reporting.
How to use it: On an eligible Xiaomi device, go to Settings > About phone > System update > Check for updates; install HyperOS 4 if the OTA is available. After updating, check whether Super XiaoAi and AI features are available for your device, region, and language.
Tip: Assume that not all AI features will be available in the global release; keep new confirmation prompts enabled for sensitive deletion or sharing actions.
Previously opened app sessions now retain the project effortLevel setting
Sessions that an app starts in advance and opens in a project no longer ignore the project’s `effortLevel` setting.
Usage details
Who can use it: App workflows that open sessions in advance, such as the Claude Desktop Code tab.
How to use it: Set `effortLevel` in the project’s settings file, start a new session in the Claude Desktop Code tab, and check the selected level.
Tip: For project-specific performance and cost targets, explicitly pin `effortLevel` in a shared settings file.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesAuto Review error before tool execution fixed on the OpenCode route
Fixed an issue that caused Auto Review to fail before a tool ran on the OpenCode route. The review request is now sent with the relevant session ID and request purpose.
Usage details
Who can use it: DeepSeek Harness v0.2.2-alpha.3; OpenCode route and Auto Review.
How to use it: Update to the latest version, start a session with the OpenCode provider, and use the Auto Review flow before a tool call; the session ID is passed automatically.
Tip: Keep the tool and Auto Review flows in the same session to associate review decisions with the correct session.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-alpha.3 sürüm notlarıPlan and provider attributes added to OpenTelemetry signals
OpenTelemetry metrics, events, and spans now include the `anthropic.plan`, `anthropic.seat_type`, `anthropic.provider`, and `anthropic.custom_base_url` attributes.
Usage details
Who can use it: Teams and administrators exporting Claude Code telemetry to OpenTelemetry.
How to use it: In your existing OTel collector, define a query or dashboard filter to receive these four `anthropic.*` attributes; send signals from the new Claude Code version.
Tip: Use these attributes as low-cardinality dimensions to separate error, cost, and usage dashboards by plan and provider.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notes`mqdh-perf capture` CLI command added for Performance Analyzer
Meta VR Developer Hub for Mac 6.6.0 adds the `mqdh-perf capture` command to capture traces from scripts with Performance Analyzer. This makes it possible to automate trace collection outside the desktop interface.
Usage details
Who can use it: Quest developers using Meta VR Developer Hub for Mac 6.6.0.
How to use it: First install Meta VR Developer Hub for Mac 6.6.0; then use the `mqdh-perf capture` command in the Performance Analyzer CLI to run your scripted trace capture workflow.
Tip: Use the same named preset in CI or repeatable performance tests to compare traces across releases.
pi-ai conversations automatically drop older images when they exceed the image limit
If a session’s image count exceeds the request limit for models connected through pi-ai, the oldest images are automatically dropped and the request continues. This prevents repeated errors caused by the image limit.
Usage details
Who can use it: DeepSeek Harness v0.2.2-alpha.3; models connected through the pi-ai provider.
How to use it: Use a session with images and a pi-ai model; if the image count exceeds the request limit, the version automatically drops older images.
Tip: In long image sessions, re-add critical images later; automatic dropping affects the oldest images.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-alpha.3 sürüm notlarıIssue creating tasks after deleting a project fixed
Fixed a bug that prevented users from creating a new task after deleting a project. The ZCode task workflow remains available after deletion.
Usage details
Who can use it: Users of the ZCode 3.17.2 desktop version.
How to use it: Update ZCode to 3.17.2; after deleting a project, use the new task button to create a task.
Tip: Before deleting a project, move any files you need to keep to another project or a safe folder.
✓ Verified by official sourceSource: ZCode Sürümler ve Güncellemeler — v3.17.2Prompt caching disable variables now apply to all default model name formats
`DISABLE_PROMPT_CACHING_OPUS` and `DISABLE_PROMPT_CACHING_SONNET` now also take effect with `[1m]`, region-prefixed, and undated default model names.
Usage details
Who can use it: Claude Code users managing prompt caching with environment variables.
How to use it: Set `DISABLE_PROMPT_CACHING_OPUS=1` or `DISABLE_PROMPT_CACHING_SONNET=1` for the relevant model and check cache behavior with different model name formats.
Tip: Normalize model names in production; verify the variable’s effect with a test request instead of assuming it’s active.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesHTTPS OpenTelemetry exports through a proxy are no longer lost
Fixed an issue where OpenTelemetry exports to an HTTPS endpoint through a proxy were never sent if they started while another export was in progress.
Usage details
Who can use it: Claude Code deployments using an HTTPS OTel endpoint behind a proxy.
How to use it: Configure an OTLP HTTPS endpoint and proxy, generate telemetry concurrently, and check that all signals reach the collector.
Tip: After upgrading, verify export delivery by tracking sequence numbers or timestamps on the collector side.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesPython SDK accumulates rich progress in Responses WebSocket
OpenAI Python SDK 3.43.0 adds support for accumulating rich progress information in the Responses WebSocket stream. This makes progress events more useful to collect client-side in long-running, multi-step Responses streams.
Usage details
Who can use it: OpenAI Python SDK v3.43.0 users; applications using Responses WebSocket.
How to use it: Upgrade with `pip install -U openai==3.43.0` and handle WebSocket events from `client.responses.connect()` in your existing event loop; read the complete response after `response.completed`.
Tip: Instead of showing only a raw event log in the user interface, turn the accumulated progress into task status and a percentage indicator.
✓ Verified by official sourceSource: OpenAI Python SDK v3.43.0 sürüm notlarıSource: Responses WebSocket modu kılavuzuPython SDK adds overflow isolation for WebSocket lanes
OpenAI Python SDK 3.43.0 adds lane overflow isolation support for Responses WebSocket. This is intended to prevent overload on one lane from affecting the progress of other lanes in clients using multiple streams.
Usage details
Who can use it: OpenAI Python SDK v3.43.0 users; applications using multiple `stream_id` values or parallel Responses WebSocket streams.
How to use it: Upgrade with `pip install -U openai==3.43.0`; run your multiple WebSocket streams with separate `stream_id` values and test each lane’s error and reconnection behavior.
Tip: Maintain separate timeouts, retries, and observability metrics for each lane; this lets you distinguish backpressure in one stream from other tasks.
✓ Verified by official sourceSource: OpenAI Python SDK v3.43.0 sürüm notlarıQoder CLI preserves compressed history when reconnecting remotely
Qoder CLI 1.1.82 fixed an issue that caused recent messages to disappear when resuming a compressed session after a remote connection was reestablished. Conversation context is expected to remain more complete after reconnecting during long sessions.
Usage details
Who can use it: Qoder CLI 1.1.82 and later; users with remote connections and compressed sessions.
How to use it: Update the CLI to version 1.1.82 or later, resume the same session through the normal resume flow after a disconnection, and check that recent messages appear in the history.
Tip: After reconnecting, verify that the last message and tool result are in the history before repeating tool calls with external effects.
✓ Verified by official sourceSource: Qoder CLI 1.1.82 resmi sürüm notlarıQoder CLI can report terminal session status via OSC 7501
Qoder CLI 1.1.82 added the ability to report session status to the terminal via the OSC 7501 protocol. Terminals or wrappers that support OSC 7501 can use these events to display or process session status.
Usage details
Who can use it: Qoder CLI 1.1.82 and terminal integrations that support OSC 7501.
How to use it: Update Qoder CLI to version 1.1.82 or later and start a session in a terminal/wrapper that can read OSC 7501 events; status events will be emitted without requiring additional settings.
Tip: If your terminal does not support OSC 7501, filter raw escape sequences or use a supporting terminal wrapper.
✓ Verified by official sourceSource: Qoder CLI 1.1.82 resmi sürüm notlarıQwen Memory Agent introduced as an open-source, measurable persistent memory system
Alibaba Cloud Community introduced Qwen Memory Agent, which runs on Qwen Cloud, as an open-source system offering persistent memory across sessions, invalidation of outdated information, and retrieval with a configurable token budget. The system provides HTTP and eight MCP tools, JSON/Markdown import and export, a live memory inspector, and a human-approved dreaming workflow.
Usage details
Who can use it: Open-source project; for developers using Qwen Cloud, Model Studio/DashScope, and Alibaba Cloud ECS. The official post provides links to the MIT-licensed source code and evaluations.
How to use it: Get the source code via the Explore the project link in the official announcement, configure a Qwen Cloud/Model Studio API key, and deploy the FastAPI backend and live memory inspector.
Tip: Keep dreaming changes subject to human approval rather than applying them automatically; start with a low retrieval token budget and adjust it to the task context.
✓ Verified by official sourceModels: Qwen Plus, qwen text-embedding-v3Source: Qwen Memory Agent duyurusuPersistent approval option removed for Remote Control and SDK mod tools
Remote Control and SDK hosts no longer mistakenly offer the `don’t ask again` option for mod tool calls.
Usage details
Who can use it: Sessions using mod tools through Remote Control or an SDK host.
How to use it: Call the mod tool in a session; use only valid options in the permission prompt and note that the persistent approval option has been removed.
Tip: If automation is needed, explicitly define permissions in the host configuration; don’t assume persistent trust from a one-time prompt.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesA recent message is answered when Remote Control restarts
Fixed an issue where, after `claude remote-control` restarted, the next message was left unanswered in an idle session if it was sent within one minute of the previous message.
Usage details
Who can use it: Users restarting Claude Code Remote Control.
How to use it: Restart Remote Control, send a message to an idle session within one minute, and check that the response arrives.
Tip: Monitor the message queue during a restart; for critical commands, confirm that the session is actually idle and connected.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesSelf-hosted runner polls using a new connection after a 5xx
After a 5xx response other than 501 or 529, the runner’s next poll now opens a new connection instead of reusing the failed connection.
Usage details
Who can use it: Organizations using self-hosted runners.
How to use it: Update the runner to v2.1.317 and check the runner logs to verify that the next poll connection is re-established after a transient 5xx.
Tip: Monitor the 5xx rate and the number of new connections; don’t rely on reconnection alone for persistent 5xx errors.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesNested session ID behavior can be reverted for self-hosted runners
`CLAUDE_CODE_DISABLE_NESTED_SESSION_ID=1` makes a `claude -p` call inside a self-hosted runner use the outer session’s ID and transcript.
Usage details
Who can use it: Teams running nested `claude -p` on a self-hosted runner.
How to use it: Add the `CLAUDE_CODE_DISABLE_NESTED_SESSION_ID=1` variable to the runner environment and run `claude -p` when you want the inner session to use the outer session’s ID/transcript.
Tip: Don’t use this variable in automations that need separate conversation histories; consider the audit implications of shared transcript behavior.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesSelf-hosted runner sessions refresh their connection after a 5xx
After a 5xx other than 501 or 529, runner sessions now open a new connection instead of using the failed one.
Usage details
Who can use it: Self-hosted runner sessions.
How to use it: Trigger a transient 5xx and monitor the session; verify in the session logs that a new connection opens automatically.
Tip: Track server error codes separately as 501, 529, and other 5xx errors; each has different recovery behavior.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesSelf-paced `/loop` wake-ups are now scheduled even when a request field is missing
Fixed an issue where a self-paced wake-up was sometimes not scheduled when Claude omitted one of the fields in a `/loop` request.
Usage details
Who can use it: Users of Claude Code’s self-paced `/loop` workflow.
How to use it: Start a self-paced task with `/loop` and check that the next wake-up time appears in the session.
Tip: For critical periodic tasks, check `/loop` status once after initial setup and add independent monitoring.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesDeleted agent team inboxes are no longer recreated by late messages
Fixed an issue where a late teammate message recreated an agent team inbox after it had been deleted.
Usage details
Who can use it: Teams using Claude Code agent teams.
How to use it: Run the agent team on v2.1.317, delete the inbox, and check that it is not recreated when a late message arrives.
Tip: Stop teammates before ending the team lifecycle; this makes late messages more predictable to manage.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesArtifact links pasted into Slack now show a simpler Open card
Newly pasted artifact links in Slack now show a simpler card with an Open button that opens the artifact in Claude, rather than the Slack side panel.
Usage details
Who can use it: Teams using Claude Tag or Claude artifact links in Slack.
How to use it: Paste a Claude artifact link into Slack; click the Open button on the card to open the artifact in Claude.
Tip: Check artifact access permissions before sharing; the recipient needs Claude access to use the Open button.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesWatermarked fields added in SpaceXAI TypeScript SDK v0.2.8
The v0.2.8 release dated November 3 added response.watermarked to Responses API responses and stream.watermarked to streams. The fields return a boolean based on the x-watermark: true header; when retryBeforeOutput is used, the header from the attempt that produced the result is used.
Usage details
Who can use it: TypeScript developers working with the Responses API using SpaceXAI TypeScript SDK v0.2.8; the SDK requires Node.js 22.13 or later and an ESM project.
How to use it: Upgrade with npm install @xai-official/sdk@0.2.8; read response.watermarked from the responses.create() result and stream.watermarked from a stream with stream: true. Enabling or disabling Watermarking is a team-level Console setting, not an SDK parameter.
Tip: Write the boolean field directly to the audit log; do not try to parse the x-watermark header separately or add this state to the request body.
✓ Verified by official sourceSource: SpaceXAI TypeScript SDK v0.2.8 sürüm notlarıSource: Watermarked alanlarının SDK commit ayrıntılarıClaude doesn’t claim success when the server refuses a deletion
Fixed an issue where Claude said the guide was deleted when the server rejected a `/team-onboarding` deletion request.
Usage details
Who can use it: Teams using server-side `/team-onboarding` guide management.
How to use it: Send a deletion request and check both the returned error and the guide list when the server rejects it.
Tip: In automations, use the server response and the actual list—not Claude’s text—as proof of deletion.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesVS Code now raises ultrathink effort when telemetry is disabled
Fixed an issue where `ultrathink` did not increase turn effort or show the rainbow indicator in VS Code when telemetry was disabled.
Usage details
Who can use it: Claude Code for VS Code users who disable telemetry.
How to use it: Disable telemetry, start a turn with `ultrathink` in VS Code chat, and check that the effort and visual indicator are enabled.
Tip: Use high effort only for difficult reasoning steps to limit unnecessary latency and usage.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notestmux and iTerm2 teammates follow the leader’s manual mode
Fixed an issue where agent team teammates in tmux or iTerm2 panes started in Auto mode while the leader session was in manual mode.
Usage details
Who can use it: Users running Claude Code agent teams with tmux or iTerm2.
How to use it: Open the leader session in manual mode, start a teammate, and check that the mode remains manual in the panes.
Tip: Set the leader’s permission mode before the team starts; don’t assume every teammate will behave automatically.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesSubcalls can be collapsed individually in the Trajectory view
Trajectory now lets you collapse subcalls independently while keeping the parent call and its summary visible. You can reopen a subcall through search or by navigating from another view.
Usage details
Who can use it: DeepSeek Harness v0.2.2-alpha.3; developer tools and the Trajectory view.
How to use it: Open a session’s Trajectory view and use the collapse control for the relevant subcall; reopen it from a search result or a cross-view link.
Tip: For long agent runs, leave open only subcalls that failed or invoked tools; this makes the summary stream faster to scan.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-alpha.3 sürüm notlarıGateway unavailability error explains network and TLS causes
When a required Claude apps gateway can’t be reached, the startup error no longer presents it as a sign-in issue; it distinguishes network access from an unverifiable TLS certificate.
Usage details
Who can use it: Claude Code installations using a required gateway or managed network policy.
How to use it: Block access to the gateway and start Claude Code; use the new error message to identify whether the cause is the network connection or TLS certificate.
Tip: Test the gateway’s DNS, firewall, proxy, and certificate chain before resetting credentials.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesParent and child skills now sync together
Fixed an issue in ZCode’s parent and child skill synchronization that caused some skills to be missed. Related skills are now updated together.
Usage details
Who can use it: Users of the ZCode 3.17.2 desktop version.
How to use it: Update ZCode to 3.17.2, start a task that uses a parent skill, and check that its linked child skills are also loaded.
Tip: After changing a skill, run a short test task to verify child skill dependencies.
✓ Verified by official sourceSource: ZCode Sürümler ve Güncellemeler — v3.17.2Stop and restart remote device services from the menu
Controls for stopping and restarting the relevant remote services have been added to the remote device menu. This reduces the need to go to the device to troubleshoot issues.
Usage details
Who can use it: ZCode v3.17.2 and later; users with a remote device connection.
How to use it: Open the device menu in the remote devices list under Settings, then select Stop service or Restart service.
Tip: Before restarting, check whether active tasks on that device could be interrupted.
✓ Verified by official sourceSource: ZCode Releases & Updates — v3.17.2Remote data synchronization consolidated in Settings
Options related to remote data synchronization have been consolidated in one place in ZCode Settings. This simplifies access to synchronization configuration.
Usage details
Who can use it: ZCode v3.17.2 and later; users with a remote session or device.
How to use it: Open the Settings page and manage the relevant options in the remote data synchronization section.
Tip: If you use multiple remote devices, check the synchronization behavior before starting a task.
✓ Verified by official sourceSource: ZCode Releases & Updates — v3.17.2Streaming response stutters reduced in long conversations
Fixed issues that caused streaming responses to stutter during long conversations. ZCode now runs more smoothly during extended response streams.
Usage details
Who can use it: Users of the ZCode 3.17.2 desktop version.
How to use it: Update ZCode to 3.17.2 and receive responses as a stream in a long conversation; no additional settings are required.
Tip: For very long tasks, archiving unnecessary history or starting a new task can help maintain smooth performance.
✓ Verified by official sourceSource: ZCode Sürümler ve Güncellemeler — v3.17.2VS Code no longer opens the sign-in screen when a network or region blocks a request
Fixed an issue where the VS Code extension opened the sign-in screen when a request was blocked due to a region or network restriction.
Usage details
Who can use it: Claude Code for VS Code users subject to region or network restrictions.
How to use it: Send a request from a blocked network and check that the extension displays an explanation of the block instead of the authentication screen.
Tip: Before repeatedly trying to sign in on a restricted network, verify proxy and regional access with your administrator.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesThe VS Code extension retries a message that was missed during startup
When a message is lost because Claude could not start in time, the extension now tries to start Claude one more time before showing an error.
Usage details
Who can use it: Claude Code for VS Code extension.
How to use it: Open the extension and reproduce the startup race; check that the second attempt preserves the message when the first startup fails.
Tip: If the second attempt also fails, record the extension logs and Claude Code version together.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesVS Code forked conversations retain parallel calls and compaction messages
Fixed an issue where forked VS Code conversations retained only one of the parallel tool calls or the final messages kept by compaction.
Usage details
Who can use it: Users who use conversation forks in Claude Code for VS Code.
How to use it: Fork a conversation in VS Code, run parallel tool calls and compaction, and check that the question and final answer are both retained.
Tip: Before forking large conversations, preserve important decisions in a short summary.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesWeChat automation result notifications are working again
Notifications for the results of WeChat automation tasks have been restored. This brings back the ability to receive results through WeChat when a task is complete.
Usage details
Who can use it: ZCode v3.17.2 and later; users of the WeChat automation integration.
How to use it: Run a WeChat automation task and check the result notification in the linked WeChat account when the task is complete.
Tip: If no notification arrives, first check WeChat authorization and the task account connection.
✓ Verified by official sourceSource: ZCode Releases & Updates — v3.17.2Incompatible components issue in the Windows 64-bit installer fixed
Fixed an issue that caused the Windows 64-bit installer to include incompatible components in some cases. Installation compatibility was improved with the 3.17.2 package.
Usage details
Who can use it: ZCode 3.17.2 Windows 64-bit installer.
How to use it: Download the 3.17.2 .exe installer for Windows from the official site and install it over the existing version.
Tip: If you encounter an issue with an older Windows installer, don’t use the same file again; download the latest official .exe package.
✓ Verified by official sourceSource: ZCode Sürümler ve Güncellemeler — v3.17.2Tool calls are no longer incorrectly rejected on some Windows volumes
Fixed an issue where every tool call was rejected on some Windows volumes when native path lookup failed; working-directory checks now compare folder identity.
Usage details
Who can use it: Claude Code users who use custom or network volumes on Windows.
How to use it: Update Claude Code to v2.1.317 and test working-directory validation by making a Read or shell call on the affected volume.
Tip: On network volumes, track actual folder identity and permissions alongside path text.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesScript errors caused by plugin command paths on Windows fixed
Fixed cases where plugin command paths caused script errors on Windows. Plugin commands now run with correct path resolution.
Usage details
Who can use it: Plugins in the ZCode 3.17.2 Windows version.
How to use it: Update ZCode to 3.17.2; run the plugin command again and, if needed, update the plugin from the official marketplace.
Tip: After updating, test plugin commands in folders with spaces in their Windows paths.
✓ Verified by official sourceSource: ZCode Sürümler ve Güncellemeler — v3.17.2Long project paths no longer push the temporary directory past 260 characters on Windows
Fixed an issue where scratchpad and temporary paths grew unnecessarily with the project path in sessions outside Claude Desktop, hitting the 260-character limit on Windows in long-path repositories.
Usage details
Who can use it: Users who open long-path repositories with Claude Code on Windows.
How to use it: Open the long-path repository with the latest Claude Code and test scratchpad and temporary file creation.
Tip: Cloning repositories to a short root path still improves compatibility with Windows tools.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesUnresponsive OTel proxies time out after 10 seconds
Fixed an issue where OpenTelemetry exports could hang indefinitely behind an unresponsive proxy; the default timeout is 10 seconds, with a maximum of 30 seconds using `OTEL_EXPORTER_OTLP_TIMEOUT`.
Usage details
Who can use it: Claude Code environments using OTel through a proxy.
How to use it: If needed, set `OTEL_EXPORTER_OTLP_TIMEOUT` to a value between 1 and 30 seconds and send a signal to the OTel endpoint.
Tip: In production, measure proxy and collector latency and choose a shorter duration rather than using the 30-second maximum.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesSelf-hosted session opens a new connection after an unanswered heartbeat
When a heartbeat goes unanswered, the self-hosted session now recovers by opening new connections instead of trying every idle connection in sequence.
Usage details
Who can use it: Self-hosted runner sessions with intermittent connection drops.
How to use it: Interrupt the heartbeat response and observe the session opening a new connection; review the runner and session logs together.
Tip: Track heartbeat timeouts and server 5xx events as separate metrics to make root-cause analysis easier.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesWarning for unintended connectors in new cloud tasks
`/schedule` now warns when a connector is added to a new cloud scheduled task without the user requesting it, and explains how to remove it.
Usage details
Who can use it: Users of the Claude Code cloud scheduled task flow.
How to use it: Create or edit a task with the `/schedule` command in Claude Code; check the summary screen for the connector warning and removal instructions.
Tip: Review the connector list before saving a scheduled task to prevent unnecessary data access.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesAn old interrupted tool call remains visible after moving to a new runner
Fixed an issue where Claude would not see an interrupted tool call when a self-hosted runner session moved to a new runner, even if the call was older than the `CLAUDE_CODE_RESUME_INTERRUPTED_TURN_MAX_AGE_MS` limit.
Usage details
Who can use it: Self-hosted environments using runner migration and interrupted turn recovery settings.
How to use it: Switch runners to resume an old interrupted turn and check that Claude sees the tool call as having an unknown outcome.
Tip: Set the age limit for automatically resuming old turns according to your operational policy; re-verify unknown outcomes.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesNew scheduled tasks use the organization’s default environment
Fixed an issue where new scheduled tasks ignored the organization’s selected default environment and started with the Anthropic-hosted default instead.
Usage details
Who can use it: Claude Code cloud scheduled tasks with an organization default cloud environment configured.
How to use it: Set the organization’s default environment in admin settings, then create a new `/schedule` task and verify the environment in the form.
Tip: Check the environment name before saving the task to prevent production secrets from being moved into a test environment.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesRecreated teammates are no longer dropped from the team due to an old shutdown approval
Fixed an issue where a recreated agent teammate was removed from its team because of an earlier shutdown approval.
Usage details
Who can use it: Claude Code agent team and teammate recreation flows.
How to use it: Recreate a teammate and connect it to the team; check that team membership is retained even when an old shutdown approval exists.
Tip: When recreating team members, track the new identity separately from the current lifecycle state.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesBuilt-in tool calls and results fully visible
The model’s process for calling built-in tools and the results of those calls are now fully shown to users. This makes it easier to monitor the task flow.
Usage details
Who can use it: ZCode v3.17.2 and later; tasks that use built-in tools.
How to use it: Run a task that uses built-in tools, then open and review the tool call and result sections in the conversation.
Tip: If you see unexpected results, check which tool the model called and what inputs it used.
✓ Verified by official sourceSource: ZCode Releases & Updates — v3.17.2Timed-out Grep and Glob results are no longer treated as complete
Fixed an issue where partial results were passed to Claude as complete when a Grep or Glob search hit its time limit.
Usage details
Who can use it: Large repositories using Claude Code Grep and Glob tools.
How to use it: Run Grep or Glob in a large repository; if it times out, narrow the search, keeping in mind that the results are partial.
Tip: Use path and file-type filters with broad patterns; don’t treat a partial search as definitive evidence.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesBulk project management and data cleanup in ZCode Settings
The Settings page now lets users manage projects in bulk. An option has also been added to delete associated data when deleting a project.
Usage details
Who can use it: ZCode v3.17.2 and later; desktop users who manage projects.
How to use it: Open project management on the Settings page, select multiple projects, and check the option to delete data when deleting them.
Tip: Before deleting, separate projects you no longer need from work data you need to keep.
✓ Verified by official sourceSource: ZCode Releases & Updates — v3.17.2Update all ZCode plugins with one click
Plugin management can now update all installed plugins in a single operation. This eliminates the need to update plugins one at a time.
Usage details
Who can use it: ZCode v3.17.2 and later; users who use the plugin manager.
How to use it: Open the plugin management or store screen and run the option to update all plugins.
Tip: Before bulk updating, finish any running tasks and note the versions of critical plugins.
✓ Verified by official sourceSource: ZCode Releases & Updates — v3.17.2Annotate ZCode image previews and send them as attachments
Image previews now support annotation and markup edits. The edited image can be sent in the conversation as an attachment.
Usage details
Who can use it: ZCode v3.17.2 and later; users who use image previews.
How to use it: Open an image in preview, use the annotation editor, and send the result in the conversation as an attachment.
Tip: When reviewing code, mark the problematic area and send it with a short instruction.
✓ Verified by official sourceSource: ZCode Releases & Updates — v3.17.2ZCode moves remote devices to a searchable compact list
Remote devices are now displayed in a compact, searchable list that takes up less space. This makes it easier to choose from many devices.
Usage details
Who can use it: ZCode v3.17.2 and later; users managing remote devices.
How to use it: Open the remote devices section in Settings and type the device name in the search field to find it.
Tip: Give devices clear names to narrow down search results faster.
✓ Verified by official sourceSource: ZCode Releases & Updates — v3.17.2Draft MCP configuration interface added to ZCode
ZCode has added a draft configuration interface for managing MCP servers. Because the interface is a draft, its behavior or fields may change in the future.
Usage details
Who can use it: ZCode v3.17.2 and later; users who use MCP. The feature is marked as a draft.
How to use it: Open the MCP configuration interface in ZCode Settings, enter the server settings, then save and test the connection.
Tip: When using the draft interface, also keep a backup of the MCP configuration in a separate file.
✓ Verified by official sourceSource: ZCode Releases & Updates — v3.17.2`/feedback` no longer incorrectly asks for confirmation while submitting
Fixed the `/feedback` screen showing `Enter to confirm` while a feedback report is being submitted.
Usage details
Who can use it: Users of the Claude Code feedback command.
How to use it: Open `/feedback` and submit a report; watch the status text while it is being submitted and don’t press Enter a second time.
Tip: Don’t close the terminal before submission completes; the updated status text indicates that the request is being processed.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notes`/schedule` results retain the next run time and link
Fixed an issue where the task’s next run time and link were missing from the results after creating or updating a cloud scheduled task.
Usage details
Who can use it: Claude Code cloud scheduled task users.
How to use it: Create or update a task with `/schedule`; open the next run time and task link in the command result.
Tip: Save the result link in your task records for operational tracking; the link now takes you directly to the next run.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notes`claude --from-pr` can find older sessions in projects with many sessions
In projects with many sessions, `claude --from-pr` no longer incorrectly reports `No conversations found` when the linked session is older than the first page.
Usage details
Who can use it: Users reopening Claude Code conversations linked to GitHub PRs.
How to use it: Update Claude Code and rerun `claude --from-pr <PR>` for the relevant PR.
Tip: In very large projects, verify the PR-to-session match using the conversation title and branch name opened after running the command.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notes`--scanning-state` removed from Agent Registry CLI
With Google Cloud CLI 592.0.0, the `--scanning-state` flag was removed from the Agent Registry skills create and update commands. Use `--scanning-policy` instead.
Usage details
Who can use it: Google Cloud CLI 592.0.0 and later; Agent Registry skill creation or update workflows.
How to use it: Use the `--scanning-policy` flag instead of `--scanning-state` in the `gcloud agent-registry skills create` and `gcloud agent-registry skills update` commands.
Tip: Update the flag in CI/CD scripts along with the version upgrade; scan your command templates and automation repositories for the old flag.
✓ Verified by official sourceSource: Google Cloud sürüm notları — Cloud SDK 592.0.0Issue causing Artifact cards not to refresh fixed
A bug that sometimes prevented Artifact cards from updating has been fixed. New or changed Artifact results now refresh more reliably in the interface.
Usage details
Who can use it: Users of the ZCode 3.17.2 desktop version.
How to use it: Update ZCode to 3.17.2; run a task that produces an Artifact and check that the card refreshes when the results are complete.
Tip: If the card does not update, close and reopen it before rerunning the task to avoid unnecessary generation usage.
✓ Verified by official sourceSource: ZCode Sürümler ve Güncellemeler — v3.17.2Settings changes made at startup are applied immediately
Fixed an issue where edits to settings files made in the first few seconds after Claude Code launched were applied only after a second change.
Usage details
Who can use it: Users who automate changes to Claude Code settings files at startup.
How to use it: Update Claude Code, change the startup settings file, and open a new session; check that the behavior is applied on the first try.
Tip: In startup scripts, leaving a short delay between writing the file and launching Claude Code can still reduce race conditions.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesSome remote connection compatibility issues fixed
ZCode 3.17.2 fixes some remote connection compatibility issues whose details are not specified in the release notes. The goal is to make Remote sessions work with different environments.
Usage details
Who can use it: ZCode Remote users on version 3.17.2; details of the affected environments are not provided in the official notes.
How to use it: Update ZCode to 3.17.2, reconnect the remote device, and reopen the project that previously had issues.
Tip: If the issue persists, include the device ID, operating system, and connection logs in your support request.
✓ Verified by official sourceSource: ZCode Sürümler ve Güncellemeler — v3.17.2Exit resume hint preserves startup flags
The `Resume this session with` hint shown on exit now repeats flags such as `--chrome` or `--debug` used to start the session, before `--resume`.
Usage details
Who can use it: Users who start Claude Code sessions with flags.
How to use it: Start a session with a flag such as `claude --chrome`, then exit; copy the displayed resume command directly.
Tip: When resuming complex sessions, copying the hint reduces the risk of forgetting required tool flags.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesThe cloud environment management dialog now uses compact sections
The organization environment dialog in admin settings now opens each section from a row, like personal dialogs, and prompts you before closing if there are unsaved entries.
Usage details
Who can use it: Organization administrators managing cloud session environments.
How to use it: Open the organization environment dialog in admin settings; edit sections from their rows, then try closing it with unsaved changes.
Tip: Read the save prompt carefully before closing; keep environment variables for production and testing in separate sections.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesCloud and SDK session titles no longer disappear when messages arrive concurrently
Fixed a race condition that caused titles to disappear in cloud and SDK sessions if a new message or Stop arrived while the title was being generated.
Usage details
Who can use it: Claude Code cloud and SDK sessions.
How to use it: Update the session; in a new session, send a message while the title is being generated and check that the title remains in the session history.
Tip: For sessions that matter for search and automation, use `/rename` after the title is generated to give it a meaningful name.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesReading an unchanged file again returns its content
The `file unchanged since your last Read` placeholder has been removed from the Read tool; reading an unchanged file again returns its content.
Usage details
Who can use it: All sessions using the Claude Code Read tool.
How to use it: Read the same file twice; check that the second call returns the file content instead of a placeholder.
Tip: Requesting the same content again consumes tokens; call Read again only when you need the current content.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesResumed session no longer repeatedly mentions a file change
Fixed an issue where Claude repeated a previously acknowledged file change in resumed sessions.
Usage details
Who can use it: Claude Code sessions resumed with `/resume` or a saved transcript.
How to use it: Resume the session, start a new turn in a project with a previously processed file change, and check that it doesn’t produce the warning again.
Tip: When you need to confirm the current file state, get a Git diff; don’t rely solely on contextual memory.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesResumed session can reuse a built-in tool
Fixed an issue where a resumed session could not reuse a built-in tool after it had previously failed because it was disabled.
Usage details
Who can use it: Claude Code users who temporarily enable or disable built-in tools and resume sessions.
How to use it: Run a turn while the tool is disabled, then enable it, resume with `/resume`, and check that the tool can be called again.
Tip: Review tool permissions before resuming a session; verify the permissions of the re-enabled tool.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesChild processes of stopped shell commands no longer receive SIGTERM twice
Fixed an issue where child processes of a stopped shell command could sometimes receive SIGTERM twice, causing programs to skip cleanup.
Usage details
Who can use it: Users who run long-running processes with the Claude Code shell tool.
How to use it: Start a long-running shell command, stop it, and check in system monitoring that child processes are cleaned up with a single stop signal.
Tip: Still design your commands to shut down safely at the application level after SIGTERM.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesPlugin settings pages support custom navigation icons
Plugins can provide a navigation icon for their settings pages. The default icon automatically returns when a plugin is removed.
Usage details
Who can use it: DeepSeek Harness v0.2.2-alpha.3; plugin developers and installed plugins.
How to use it: Add the navigation icon to your plugin’s settings page definition and update Harness; check that the default icon returns when the plugin is removed.
Tip: Use a consistent icon style across a plugin family; customize an icon only when it genuinely helps distinguish the settings page’s function.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-alpha.3 sürüm notlarıOld transcripts are also cleaned up with custom setting sources
Fixed an issue where old transcripts weren’t cleaned up in some runs using `--setting-sources`.
Usage details
Who can use it: Claude Code sessions using `--setting-sources` or similar setting sources.
How to use it: Update Claude Code and start a session with `--setting-sources`; verify that old transcripts are cleaned up by monitoring the Claude directory.
Tip: If you need a transcript archive, set up a separate archiving policy before relying on automatic cleanup.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesA claude.ai connector in a matching plugin no longer asks for unnecessary authorization
Fixed a misleading notification claiming that the claude.ai connector running in a cloud session needed reauthorization or had failed when a matching plugin contained the same server.
Usage details
Who can use it: Users of claude.ai connector and matching plugins in cloud sessions.
How to use it: Match the plugin and use the connector in a cloud session; check that the authorization warning disappears when the server is already matched.
Tip: Prefer a single source instead of adding the same MCP server as both a connector and a plugin.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesGemini 4 Argon reaches general availability in Snowflake AI Functions
According to Snowflake’s official release notes, the `gemini-4-argon` model is GA in AI Functions and has been added to regional and cross-region usage options. Since I could not find separate confirmation of this Snowflake access through Google’s own channels, the entry is marked `verified=false`.
Usage details
Who can use it: Snowflake AI Functions; regional and cross-region availability depends on Snowflake’s Cortex AI Functions model table.
How to use it: Select `gemini-4-argon` as the model in Snowflake AI Functions and check the relevant regional or cross-region support table before making a call.
Tip: Before going to production, verify model access, quota, and pricing separately for the region you selected in Snowflake.
! UnverifiedModels: Gemini 4 ArgonSource: Snowflake sürüm notu — 3 Kasım 2026: Google Gemini 4 Argon GAGemini CLI no longer treats interrupted Whisper downloads as installed
The 2026-11-03 nightly release fixes an issue that incorrectly treated interrupted Whisper model downloads as installed. This prevents a corrupted or incomplete download from being skipped as if it had completed successfully.
Usage details
Who can use it: Gemini CLI nightly channel; version v0.67.0-nightly.20261103.g0ac57c029.
How to use it: Update Gemini CLI nightly to v0.67.0-nightly.20261103.g0ac57c029 or later, then restart the interrupted Whisper download.
Tip: After network interruptions, keep the nightly release up to date and check that the model is actually available before the first call.
✓ Verified by official sourceModels: WhisperSource: Gemini CLI v0.67.0-nightly.20261103 GitHub sürümüBug fixes released for Gemini in VS Code
The Gemini section of Google Cloud’s November 3 release notes lists various bug fixes and minor product improvements for VS Code. The notes do not detail the fixes individually or specify a version number.
Usage details
Who can use it: Developers using Gemini for Google Cloud in VS Code; deployment and extension version are unspecified.
How to use it: Update the Gemini for Google Cloud extension from the VS Code Extensions panel and reload the window if prompted.
Tip: When reporting an issue, first update the extension to the latest version and restart VS Code; this makes it easier to rule out problems caused by an outdated client.
✓ Verified by official sourceSource: Google Cloud sürüm notlarıThe rollout of creating custom avatars from images in Google Vids has been paused
The rollout of the feature for creating custom avatars in Google Vids by uploading an image from a local computer, Drive, or Photos was paused on November 3. Google did not provide a restart date and updated the feature’s rollout timeline to TBD.
Usage details
Who can use it: The rollout of the feature to the Business, Enterprise, Education Plus, personal Google accounts, and relevant add-on plans listed in the announcement is paused.
How to use it: You can’t enable this feature right now; when the rollout resumes, open the avatar creation flow in Google Vids and use the image upload option.
Tip: Plan team training and content templates after the rollout resumes; in the meantime, use Vids’ existing flow for creating avatars from scratch.
✓ Verified by official sourceSource: Google Workspace Updates — Google Vids özel avatar güncellemesiNested `claude -p` no longer sends its own title to the outer session
Fixed an issue where a `claude -p` call inside a Cloud or self-hosted runner generated its own title and sent it to the outer session.
Usage details
Who can use it: Users running nested `claude -p` calls inside runner sessions.
How to use it: Run the nested call and check the outer session history to confirm that only the outer session’s title stream remains.
Tip: Instead of relying on automatic title generation, explicitly rename important outer sessions afterward.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesExplanation with the setting name shown when Remote Control is off
When Remote Control is off, `/remote-control` no longer says `Unknown command`; it explains that the `disableRemoteControl` setting disables the feature.
Usage details
Who can use it: Claude Code installations using the `disableRemoteControl` setting.
How to use it: Turn off Remote Control and run `/remote-control`; enable the feature by turning on the `disableRemoteControl` setting mentioned in the message.
Tip: If your organization’s policy turns off Remote Control, show users an explanation of the setting and policy instead of a command.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesChinese and English resource issue in the macOS installer fixed
Fixed an issue that caused the macOS installer to omit Chinese and English language resources in some installations. Language resources are expected to be available in new installations.
Usage details
Who can use it: ZCode 3.17.2 macOS installers; Apple Silicon and Intel packages.
How to use it: Download the Mac package for 3.17.2 from the official ZCode download page, then update or reinstall your existing installation.
Tip: If the installation interface appears incomplete, download the official installer again instead of using a third-party package.
✓ Verified by official sourceSource: ZCode Sürümler ve Güncellemeler — v3.17.2Issue causing blank app icons on macOS fixed
A bug that caused some ZCode app icons to appear blank on macOS has been fixed. Icons are expected to display correctly in an up-to-date installation.
Usage details
Who can use it: ZCode 3.17.2 for macOS; Apple Silicon and Intel packages.
How to use it: Update the official ZCode package to 3.17.2; if the icon cache is causing an issue, close and reopen the app.
Tip: If the icon remains blank after updating, reinstall the official package rather than replacing the app with a third-party copy.
✓ Verified by official sourceSource: ZCode Sürümler ve Güncellemeler — v3.17.2Menus and floating panels now stay within the visible area
Menus and floating panels track browser zoom, scrolling, and on-screen keyboard changes to stay within the visible area as much as possible. This improves accessibility, especially on touch devices and small screens.
Usage details
Who can use it: DeepSeek Harness v0.2.2-alpha.3; web interface.
How to use it: Update to the new version; open a panel in a window with zoom enabled, scrolled, or with the on-screen keyboard open to use automatic positioning.
Tip: If a panel is partly off-screen on mobile, close and reopen it instead of resizing the window; the new layout adapts to the visible area.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-alpha.3 sürüm notlarıText and Markdown previews preserve reading position after refresh
Text and Markdown file previews now preserve the current reading position after a refresh. Even if the file temporarily gets shorter, the view no longer unnecessarily jumps back to the beginning.
Usage details
Who can use it: DeepSeek Harness v0.2.2-alpha.3; text and Markdown previews.
How to use it: Open a file in preview, scroll down, then refresh or update the file; the preview keeps your place.
Tip: Refresh while monitoring long log or output files without having to scan for changes from the beginning each time.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-alpha.3 sürüm notlarıMiMo Code history search avoids fetching unnecessary index rows
History search now reads only the requested number of rows, since there is one index row per chunk; the previous limit×24 prefetch behavior has been removed. In Xiaomi’s test on a 47,000-row database, the latency of queries with a limit of 50 fell from approximately 69–114 ms to 8–27 ms, while result ordering was preserved.
Usage details
Who can use it: The history search engine on MiMo Code’s GitHub main branch; especially effective for searches running on Desktop’s main thread.
How to use it: Use MiMo Code history search through the existing search interface or the History.search function, specifying the `limit` value you need; the engine now reads according to that limit.
Tip: For large histories, request only the number of results you need with `limit`; this keeps search wait times low.
✓ Verified by official sourceSource: MiMo Code GitHub commit: stop over-fetching history search rowsPreview cards now open when you click anywhere on the card
Preview cards can now be opened by clicking anywhere on the card, rather than only on a specific link.
Usage details
Who can use it: ZCode v3.17.2 and later; users who see preview cards.
How to use it: Click anywhere on a preview card to open it.
Tip: Click directly on the card body instead of looking for the small link.
✓ Verified by official sourceSource: ZCode Releases & Updates — v3.17.2Session updates and long-response streaming use less processing overhead
Alpha.3 reduces duplicate emissions and catalog operations in session state updates. Streaming and assembling long responses also incurs lower runtime overhead.
Usage details
Who can use it: DeepSeek Harness v0.2.2-alpha.3; long responses and intensive session states.
How to use it: Update to the new version and run a long streaming response or a session with many updates; the performance improvement is applied automatically.
Tip: Reduce unnecessary parallel session emissions during long tasks; compare streaming latency on the same task after updating.
✓ Verified by official sourceSource: DeepSeek Harness v0.2.2-alpha.3 sürüm notlarıPDF metadata can no longer throw off page counts in Read and mentions
Fixed an issue where a PDF’s own metadata caused the page count to be underestimated in Read and `@` mention operations.
Usage details
Who can use it: Users working with PDF files via the Claude Code Read tool or mentions.
How to use it: Open a PDF containing metadata with the Read tool or attach it with `@`; check that all pages are counted.
Tip: For very long PDFs, specify the page range explicitly to manage both accuracy and token usage.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notes“GPU Surface Timing” preset removed from Performance Analyzer
Meta VR Developer Hub for Mac 6.6.0 removes the legacy “GPU Surface Timing” (Standard VR) preset. Trace capture now falls back to the CPU Low Overhead preset instead.
Usage details
Who can use it: Quest developers using Meta VR Developer Hub for Mac 6.6.0; affects legacy GPU Surface Timing workflows.
How to use it: Move commands and scripts tied to GPU Surface Timing to the CPU Low Overhead preset, then test the new trace capture workflow.
Tip: Do not treat measurements from the old and new presets as equivalent in performance comparisons; take a new baseline after migrating.
Exiting no longer prompts for background work approval when a plugin monitor is running
Claude Code now skips the `Background work is running` confirmation when exiting while plugin monitors are running.
Usage details
Who can use it: Claude Code sessions using plugin monitors.
How to use it: Exit Claude Code while a plugin monitor is running; observe the direct exit behavior instead of the usual background work confirmation.
Tip: Manage the monitor’s lifecycle separately; skipping the exit confirmation may not automatically stop persistent monitoring.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesQoder workspace and recent task lists no longer get stuck on refresh
Qoder 0.5.3 fixed an issue that left workspace and recent task lists in an endless loading state during repeated updates. The goal is to make the list view more stable during refreshes.
Usage details
Who can use it: Qoder 0.5.3 and later; users of workspace and recent task lists.
How to use it: Update Qoder to version 0.5.3 or later, then reopen and refresh the workspace or recent task list.
Tip: Instead of refreshing the list repeatedly, wait for the status indicator to finish after a single refresh request.
✓ Verified by official sourceSource: Qoder 0.5.3 resmi sürüm notlarıQoder can load old tasks with empty source session fields
Qoder 0.5.3 fixed an issue that prevented the task list from loading because some past conversations had an empty source session value. These old records are now handled more safely instead of disrupting the list of valid tasks.
Usage details
Who can use it: Qoder 0.5.3 and later; users with past conversation and task records.
How to use it: Upgrade Qoder to version 0.5.3 or later and reopen the past task list.
Tip: On installations with historical data, update all devices to the same version; different client versions can make comparisons difficult.
✓ Verified by official sourceSource: Qoder 0.5.3 resmi sürüm notlarıRemote Control fork names now distinguish conflicts more clearly
When the application doesn’t provide a fork name, the `(fork)` suffix is preserved for long names; a new fork from a `Name (fork)` session is named `Name (fork 2)`.
Usage details
Who can use it: `claude remote-control` and application-based Remote Control fork sessions.
How to use it: Fork a Remote Control session; check the new name for long names and names that have already been forked.
Tip: To distinguish sessions, explicitly rename them after forking to include the job or branch.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesBrowser attachments are no longer downloaded repeatedly
A bug that caused attachments in the ZCode browser to be downloaded repeatedly for the same content has been fixed. This reduces unnecessary downloads and waiting time.
Usage details
Who can use it: Users of the built-in browser in ZCode 3.17.2.
How to use it: Update ZCode to 3.17.2; reopen an attachment link in the built-in browser to check that it downloads only once.
Tip: For large attachments, monitor the download folder to verify that duplicate copies of the same file are not created.
✓ Verified by official sourceSource: ZCode Sürümler ve Güncellemeler — v3.17.2The VS Code Account panel shows the plan name with correct capitalization
Fixed an issue where the second word of the plan name appeared in lowercase in the VS Code Account panel.
Usage details
Who can use it: Claude Code for VS Code Account panel.
How to use it: Open the Claude Account panel in VS Code and check that the plan name appears correctly.
Tip: Don’t treat the displayed plan name as the sole proof of access issues; check model and organization access separately.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesIncreasing VS Code `chat.fontSize` also scales panel components
Fixed an issue where attachment chips, menus, banners, and dialogs remained small when `chat.fontSize` was increased.
Usage details
Who can use it: Claude Code users who increase the `chat.fontSize` setting in VS Code.
How to use it: Increase `chat.fontSize` in the VS Code `settings.json` file and check that all components in the Claude panel scale accordingly.
Tip: For accessibility, increase the panel width as well as the text size to reduce wrapping in long prompts.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesVS Code file links open in a separate editor group
Fixed an issue where file links in the Claude tab opened over the chat; links now open in another editor group.
Usage details
Who can use it: Claude Code for VS Code chat panel.
How to use it: Click a file link in Claude’s response and check that the file opens in a separate editor group without closing the chat.
Tip: Use VS Code’s editor group layout to view code and chat side by side.
✓ Verified by official sourceSource: Claude Code v2.1.317 release notesWindows window displays correctly after being covered and uncovered
Fixed an issue that caused the ZCode window to display incorrectly on Windows after being covered by another window. The interface is drawn correctly when the window becomes visible again.
Usage details
Who can use it: ZCode 3.17.2 Windows version.
How to use it: Update ZCode to 3.17.2; cover the window with another window, then bring it to the foreground again to check the display.
Tip: If the display is corrupted, try minimizing and restoring the window before restarting the app.
✓ Verified by official sourceSource: ZCode Sürümler ve Güncellemeler — v3.17.2Files open with the default app on Windows
Fixed a bug that prevented some files from opening with the default app on Windows. File cards and the open menu now work more consistently.
Usage details
Who can use it: ZCode 3.17.2 Windows version.
How to use it: Update ZCode to 3.17.2; use the open menu on a file card to open the file in the default app.
Tip: If the issue persists, check the file type association in Windows Settings > Apps > Default apps.
✓ Verified by official sourceSource: ZCode Sürümler ve Güncellemeler — v3.17.2Unread filter for ZCode tasks
A new filter was added to show only unread tasks in the task list.
Usage details
Who can use it: ZCode v3.17.2 and later; all task list users.
How to use it: Open the Tasks view and select the Unread filter from the filter options.
Tip: Use this filter to quickly scan notifications for work completed in the background.
✓ Verified by official sourceSource: ZCode Releases & Updates — v3.17.2ZCode tray badge removed; unread status remains on the taskbar
The unread red badge is no longer shown on the ZCode system tray icon. Unread status remains visible on the Windows taskbar.
Usage details
Who can use it: ZCode v3.17.2 and later; Windows desktop users.
How to use it: Track tasks completed in the background using the unread status on the Windows taskbar; don’t look for a badge in the system tray.
Tip: Pin ZCode to the taskbar and keep taskbar badges enabled so you don’t miss notifications.
✓ Verified by official sourceSource: ZCode Releases & Updates — v3.17.2ZCode installer size significantly reduced
The installer size has been significantly reduced in ZCode 3.17.2. This lowers the amount of data required for download and installation.
Usage details
Who can use it: Reportedly applies to ZCode 3.17.2 desktop packages for macOS, Windows, and Linux.
How to use it: Download the 3.17.2 installer for your operating system from the official ZCode download page.
Tip: On a limited connection, save the official package locally to avoid interrupting the download.
✓ Verified by official sourceSource: ZCode Sürümler ve Güncellemeler — v3.17.2Create plugins from chat with Creator Mode in DeepSeek Harness
DeepSeek Harness introduced Creator Mode, which turns ideas described in natural language into usable plugins. The official announcement showcases examples such as a design workspace, a sandbox game, and a 3D LLM visualizer.
Usage details
Who can use it: DeepSeek Harness users; offered as a worldwide public preview on the product page.
How to use it: Open DeepSeek Harness, select Creator mode in a new session, and describe the plugin you want to create in natural language; install and validate the resulting plugin.
Tip: In your initial request, define the plugin’s purpose, interface, and security boundaries together; check the generated files and permissions in the Review view before installing.
✓ Verified by official sourceSource: DeepSeek’in Creator Mode duyurusuSource: DeepSeek Harness resmî ürün sayfasıGrok Bot’s phone-calling feature is rolling out gradually to accounts
An official @grok reply dated November 2 said that Grok Bot’s contact-calling flow was gradually rolling out to accounts using Grok Bot 0.77+ on the iOS app and desktop. The bot can call the requested contact, provide a live transcript during the call, and deliver a summary afterward; no additional add-on is required.
Usage details
Who can use it: iOS app and desktop Grok Bot 0.77+; account rollout is gradual, with plan and country availability unspecified.
How to use it: Update Grok Bot on iOS or desktop to version 0.77+; once the feature is enabled on your account, write in the chat who to call and what you need.
Tip: In one message, state the purpose of the call, the information to convey to the other person, and the desired outcome; after the live transcript, review the summary and actions taken.
✓ Verified by official sourceSource: Grok'un X yanıtı: telefon araması, canlı döküm ve özetgrok-imagine-image-quality retired; requests routed to Imagine 2.0
xAI retired the grok-imagine-image-quality model ID from the API on 2 Kasım; the old ID continues to work, but requests are served by grok-imagine-image-2.0 with quality: low. The request and response formats remain unchanged; this routing is 0,01 ABD doları cheaper per image than the previous model.
Usage details
Who can use it: All requests using the old ID on the /v1/images/generations and /v1/images/edits endpoints in the xAI API are affected. grok-imagine-image 1.0 is not affected; grok-imagine-image-pro also routes to Imagine 2.0 low through the chain.
How to use it: Set the model field in your code to grok-imagine-image-2.0 and choose low, medium, or auto for quality; if you continue using the old ID, low is applied automatically. Log the model field in the response to verify the transition.
Tip: If you want to lock in the cost and output, specify the quality value explicitly; if quality matters, use medium and call the new model directly rather than relying on the old ID in production.
✓ Verified by official sourceModels: grok-imagine-image-quality, grok-imagine-image-2.0, grok-imagine-image-proSource: grok-imagine-image-quality emekliliği için geçiş kılavuzuSource: xAI API sürüm notlarıMods prompt.submit hooks now run in headless executions
In headless -p and stream-json requests, Mods prompt.submit hooks can now modify the request, add hidden context, or drop the request before the model sees it. Failing hooks do not block the flow by default; terminal paths are out of scope for this release.
Usage details
Who can use it: Headless Qwen Code CLI runs; experimental.mods must be enabled and the relevant Mod must be installed.
How to use it: Install the Mod with `qwen extensions install ./guard-mod --consent`, enable experimental.mods, and use `qwen -p 'prompt' --output-format json` or stream-json.
Tip: Test a guard Mod that prevents credential leaks under an isolated HOME first.
✓ Verified by official sourceSource: Qwen Code v0.26.6 resmî sürüm notlarıSource: Headless prompt.submit kancaları değişikliğinin ayrıntılarıKimi Open Platform launches Admin API for organization management
Kimi Open Platform has launched the Admin API in beta, enabling verified organizations to programmatically manage members, invitations, projects, and inference API keys, and query billing, usage, and audit logs. The API is free and operates as a control plane; Admin API keys cannot be used for model inference calls.
Usage details
Who can use it: Kimi Open Platform organizations that have completed organization verification and organization administrators; beta.
How to use it: Complete verification under Console → Organization → Organization Verification. Then create a key with the required scopes under Console → Organization → Admin API Keys; set the key as the MOONSHOT_ADMIN_KEY environment variable and send a request to the api.moonshot.ai/v1/admin/organizations endpoint with Bearer authorization.
Tip: Grant only the scopes you need: org:write for write operations, billing:read for billing and usage, and audit:read for audit logs. Don’t confuse the Admin API key with the inference API key.
Message Threads beta arrives in the Messages API for long sessions
The Messages API now supports the Message Threads beta, which temporarily stores conversations on the server: the first request sends the full context, while subsequent requests send only new messages. Anthropic says median time to first token for 500,000–1 million-token requests is about 35 percent lower, but token costs and rate limits are unchanged.
Usage details
Who can use it: Claude API, Claude Platform on AWS, Anthropic-managed Claude Models on Google Cloud, and Microsoft Foundry; beta on all supported models. A thread ends after five minutes of inactivity and is deleted no later than one hour after its last use; the feature is not eligible for ZDR/HIPAA.
How to use it: Add the `anthropic-beta: message-threads-2026-08-12` header to a Messages API request. In the first request, send `thread: {type: create}` and the full `messages`, `system`, and `tools`; in subsequent requests, send only new messages with `thread: {type: continue, previous_message_id: ...}`.
Tip: Use threads in active, long-running agent loops; keep a fallback in your client that can reconstruct the full conversation, since a thread may expire after five minutes of inactivity.
✓ Verified by official sourceModels: Claude Opus 5.5Source: Make long-running agents faster with message threadsSource: Message threads documentationSource: Claude Platform release notesMiMo Code reverts the project configuration execution boundary on the main branch
The official revert dated November 2 rolled back the host-source boundary added on October 30; the hosted Node engine no longer automatically filters MCP, LSP, and formatter sections in project configuration. Project config can therefore provide these executable settings again; this is a later rollback change, distinct from the October 30 feature in the known list.
Usage details
Who can use it: MiMo Code installations using the hosted Node engine and the XiaomiMiMo/MiMo-Code main branch. The change was merged into main on November 2, 2026; no separate release tag has been published since November 1.
How to use it: Update from the main branch and review project config files again in the hosted Node environment. Remove untrusted MCP, LSP, and formatter settings; you can use the MIMOCODE_ONLY_CONFIG_DIR_PLUGINS variable to limit only the plugin scope to the config directory.
Tip: When running untrusted projects on the hosted Node engine, restrict MCP, LSP, and formatter sources to trusted repositories; project configuration no longer enforces this boundary automatically.
Qoder Cloud Agents introduces a continuous evaluation loop
Qoder Cloud Agents introduced a continuous agent evaluation framework that uses the actual execution environment and evaluates traces, tool calls, responses, and outputs together. The release approach covers regression checks using isolated test sessions, baseline comparisons, custom datasets, and copies that do not affect production.
Usage details
Who can use it: Qoder Cloud Agents users; additional evaluation capabilities are expected to roll out gradually.
How to use it: Create an evaluation dataset at https://qoder.com/cloud/quickstart, run representative cases, and compare the results from agent copies with the baseline.
Tip: Include real production failures in the initial dataset, and check not only the final response but also the process and generated files.
✓ Verified by official sourceSource: Qoder Cloud Agents sürekli değerlendirme çerçevesiThe Web Shell GitHub Channel setting becomes a full GitHub form
GitHub Channel configuration now offers a labeled checklist of notification reasons, an explicit choice between a token and local gh, and owner/repo validation. The repository allowlist uses case-insensitive matching, and custom message fields that are not specific to GitHub have been removed from the form.
Usage details
Who can use it: Administrators and developers using GitHub Channel in Web Shell; GitHub Enterprise base URLs are also supported.
How to use it: Go to Settings → Channels → Configure GitHub; complete the Credentials, Connection, Notifications and routing, and Access control sections, then click Save.
Tip: If you choose local gh authentication, check that the saved token will be removed; enter allowlist items directly in owner/repo format.
✓ Verified by official sourceSource: Qwen Code v0.26.6 resmî sürüm notlarıSource: GitHub Channel yapılandırması değişikliğinin ayrıntılarıverify-pr rechecks in the same session retain --publish approval
verify-pr rechecks in the same session now inherit the previous --publish approval. Users can publish ongoing verification rounds without repeatedly granting the same publishing permission.
Usage details
Who can use it: The same sessions using Qwen Code verify-pr workflows.
How to use it: Run the first round with the ` /verify-pr --publish` command; when you start a recheck in the same session, it continues without asking for approval again.
Tip: Do not use --publish in the first round if you do not want to generate a publishable report; approval is inherited for the session.
✓ Verified by official sourceSource: Qwen Code v0.26.6 resmî sürüm notlarıPer-folder Git and PR status comes to multi-directory workspaces
Kimi Code 2.5.0-canary now shows Git status, change statistics, file diffs, and PR information separately for each folder in multi-directory workspaces. Results and errors are distinguished for each folder in batch queries; draft PRs can now be returned with draft status.
Usage details
Who can use it: Multi-directory workspaces in Kimi Code Desktop and Web; normal single-directory behavior is unchanged.
How to use it: Open a workspace containing multiple source folders in Kimi Code Desktop or Web, then open the Changes view in the session’s right panel; the panel loads Git and PR status separately for each root folder.
Tip: Organize folders in the same workspace as separate repositories or subdirectories; if one folder has a Git error, the status of the other folders will still be shown.
Experimental unifiedAgentLoop brings more ACP turns into the shared loop
With experimental.unifiedAgentLoop enabled, channel prompts, Goal suggestion approvals, retry, continue, and ask_user_question restore turns run through a shared model turn and batched tool execution. This lets different ACP paths share the same stopping and budget rules.
Usage details
Who can use it: Qwen Code sessions using ACP; the feature is off by default and requires an experimental setting.
How to use it: Set experimental.unifiedAgentLoop to true in your settings, then run a channel prompt, retry, continue, or question restoration flow through ACP.
Tip: Run your parity tests before using this in production; behavior is unchanged when the setting is off.
✓ Verified by official sourceSource: Qwen Code v0.26.6 resmî sürüm notlarıSource: ACP ortak döngü kapsamı değişikliğinin ayrıntılarıFedora 44 and RHEL 10 RPMs added to Desktop Linux packages
In addition to existing Linux packages, Qwen Code Desktop releases RPM installers for x64 Fedora 44 and RHEL 10. This expands package-manager installation options on these distributions.
Usage details
Who can use it: x64 Linux desktop users running Fedora 44 and RHEL 10.
How to use it: Download the RPM file for your operating system from the Qwen Code v0.26.6 GitHub Releases page and install it with `sudo dnf install ./dosya.rpm`.
Tip: After installation, verify Desktop's daemon PATH setting and the package update channel.
✓ Verified by official sourceSource: Qwen Code v0.26.6 resmî sürüm notlarıOriginal paths and parent folders are shown for invalid artifact links
When an artifact link is invalid, Kimi Work can now display the original file path. You can continue looking for the file by opening the parent folder from the same screen.
Usage details
Who can use it: Kimi Work desktop app versions 3.2.24 and later.
How to use it: Open an invalid artifact link; check the original path shown and use the option to open the parent folder.
Tip: Noting the original path before deleting or moving output folders makes it easier to recover files from old artifact links.
✓ Verified by official sourceSource: Kimi Work Release NotesGLM-5.3 enters production at T-Systems Industrial AI Cloud in Munich
Z.ai announced that GLM-5.3 is in production on T-Systems’ Industrial AI Cloud in Munich. The model is now available to European enterprise customers in Germany with local hosting and data sovereignty options.
Usage details
Who can use it: T-Systems AI Foundation Services users; in Germany on Industrial AI Cloud/T-Cloud Public, especially in Munich. The catalog lists Essential, Professional, and Agentic plans.
How to use it: Get a trial key from the T-Systems API Key Portal (for production, use T-Cloud Marketplace); set OPENAI_BASE_URL to https://llm-server.llmhub.t-systems.net/v2 and send model: "GLM-5.3" in a /chat/completions request.
Tip: For projects requiring data residency in Germany or the EU, verify T-Cloud hosting and the 1M context window; also compare GLM-5.3-Flash for lower cost or image input.
Goal approval text is sanitized for control and bidirectional characters
Web Shell Goal approval text and edit diff file paths are now displayed with control and bidirectional Unicode characters escaped. This reduces the risk of misleading users through hidden directional or control characters in visible content.
Usage details
Who can use it: All sessions displaying Goal approvals and edit diffs in Web Shell.
How to use it: Open a Goal or edit approval and check that the text and file paths are displayed using visible characters.
Tip: Before approving, copy the file path and inspect its raw value too; the displayed text may differ from the actual path.
✓ Verified by official sourceSource: Qwen Code v0.26.6 resmî sürüm notlarıGoogle Antigravity: Removal date arrives for Claude 4.6 and GPT-OSS 120B
Google Antigravity’s official model table says Claude Sonnet 4.6, Claude Opus 4.6, and GPT-OSS 120B will be removed on November 2, 2026. You need to move workflows using these models to Gemini 3.8 Flash, which is available for your plan in Antigravity, or to Gemini 4 Argon if you have an eligible paid Google AI Pro or Ultra account.
Usage details
Who can use it: Affects model selection in Antigravity App, Antigravity CLI, and Antigravity IDE. Claude Sonnet 4.6 and Claude Opus 4.6 were listed in the free/Plus and Pro tables, while GPT-OSS 120B was listed in the free/Plus, Pro, and Ultra tables; none of these three models are listed for Enterprise.
How to use it: In Antigravity, select Gemini 3.8 Flash from the model picker below the chat prompt. If you have a paid Google AI Pro or Ultra account, switch to Gemini 4 Argon from the same picker and retest important tasks with the new model.
Tip: Before the removal date, run a comparison test on Gemini 3.8 Flash for your critical prompts and simplify model-specific instructions; this helps reduce sudden behavior differences in production workflows.
✓ Verified by official sourceModels: Claude Sonnet 4.6, Claude Opus 4.6, GPT-OSS 120BSource: Google Antigravity Docs: Model seçimiCreate a custom avatar in Google Vids by uploading an image
Google Vids now lets you create a custom AI avatar from an image uploaded from your computer, Drive, or Photos. You can refine the image in Vids and reuse the avatar within the same project.
Usage details
Who can use it: Visibility in Rapid Release domains is rolling out starting October 30, 2026; full rollout in Scheduled Release domains begins November 20, 2026. Available to Business Starter/Standard/Plus, Enterprise Starter/Standard/Plus, Education Plus, personal Google accounts, Google AI Pro and Ultra, Enterprise Essentials/Essentials Plus, Nonprofits, Individual, Google AI Pro for Education, Teaching and Learning, and AI Expanded Access.
How to use it: Open Google Vids > AI avatar > Design a custom avatar > Upload image > Upload. Select an image, edit it with Refine selected if needed, add a voice and name, then click Save > Select; enter your text and select Insert.
Tip: For best results, use a well-lit, 16:9 landscape image with a clear face and neutral background; you can reuse the avatar stored in the project in the same video.
Shared Data Bot data science agent introduced in Grok Bot
xAI announced Data Bot, a Team Bot trained by its data science team. Data Bot can query the data warehouse and generate analyses and charts/reports; teams can add it to Slack channels and receive scheduled reports.
Usage details
Who can use it: The announcement presents Data Bot for team use within xAI/SpaceXSI; it does not specify a public ready-made template or country/plan coverage. It offers an example use case for teams with the general Grok Bot Team Bot and Slack capabilities.
How to use it: The announcement does not offer a public one-click setup for Data Bot. To build the same setup, select + → Create a Bot in Grok Bot, define the data role, sources, and approval boundaries in the first message; authorize your workspace via Connect apps → Slack and add the Bot to the relevant channel.
Tip: Clearly document metric definitions, query logic, and data owners in shared skills; require read-only access to the data store and approval for external actions.
Fine-tuned Gemini 3.x models reach GA on global endpoints
Gemini Enterprise Agent Platform has made serving fine-tuned Gemini 3.x models on global endpoints Generally Available (GA). Developers can now copy fine-tuned models to the global region through Model Registry and deploy them to a global serving endpoint.
Usage details
Who can use it: Generally Available (GA) for projects fine-tuning Gemini 3.x models on Gemini Enterprise Agent Platform; the global region and the direct `aiplatform.googleapis.com` endpoint are available.
How to use it: Copy the fine-tuned Gemini 3.x model to the `global` region using the Model Registry `models:copy` API, deploy it to a global serving endpoint, and send inference requests through `aiplatform.googleapis.com`.
Tip: The global endpoint expands access; for workloads with data residency or compliance requirements, compare your needs against the `us` and `eu` multi-region options before switching to global deployment.
✓ Verified by official sourceModels: Gemini 3.5 Flash-Lite, Gemini 3.5 Flash, Gemini 3.1 Flash-LiteSource: Gemini Enterprise Agent Platform sürüm notları: tuned Gemini 3.x modellerinin global uç noktalarda GA olmasıSource: Google Cloud: Gemini modelleri için supervised fine-tuning dokümantasyonuChannel users can approve their own one-time tool permissions
Channel users can now send `/approve` or `/deny` for their own pending one-time tool permissions in the current attended turn. This lets them complete the permission decision through the channel without returning to the terminal.
Usage details
Who can use it: Users with attended turns through Qwen Code channel adapters.
How to use it: When a pending one-time tool permission request arrives, send `/approve` or `/deny` in the same channel.
Tip: Use these one-time commands when you want to approve only the current request rather than set a permanent permission policy.
✓ Verified by official sourceSource: Qwen Code v0.26.6 resmî sürüm notlarıKimi Code combines the Skills catalogs of all workspace directories
Kimi Code 2.5.0-canary now discovers, lists, and invokes Agent Skills from every configured source directory in a multi-directory workspace through a single catalog. Changes in directories are monitored and reloaded; for Skills with the same name, the primary directory and workspace order determine which one takes precedence.
Usage details
Who can use it: Multi-directory CLI, Desktop, Web, TUI, and SDK workspaces using the Kimi Code 2.5.0-canary preview.
How to use it: Add top-level extra_skill_dirs = ["~/team-skills", ".agents/team-skills"] to the ~/.kimi-code/config.toml file, or add multiple source directories to the workspace; leave the merge_all_available_skills setting as true and start a new session. To invoke a Skill manually, use /skill:<skill-adı>.
Tip: If you use Skills with the same name, arrange the primary working directory and workspace order deliberately; temporary directories added only to a session won’t appear in the shared catalog, so you need to register them as permanent source directories.
Kimi Memory Space: Users report new memory behavior
A November 2 report by 36Kr says some users have reported in recent days that Memory Space would not load and that the new version consolidates the old, individual memory instructions into summaries of user preferences. Moonshot has not officially announced this specific behavior change on November 2; its official Help Center only confirms that the new Memory experience has been rolling out gradually since October 2026.
Usage details
Who can use it: On some Kimi accounts; according to the official Help Center, the interface and settings may vary by platform, app version, and region.
How to use it: In Kimi, open Settings → Personalization → Memory or Memory Space; view, export, or delete saved memories. If the new section is not visible, the rollout may not yet be complete for your account.
Tip: Check the source indicator in responses that use memory; if you see an outdated or incorrect memory, delete it from Memory Space and save the updated information again.
Kimi Work network settings automatically follow the system proxy
With Kimi Work 3.2.24, network settings dynamically follow the system proxy configuration. When the system proxy changes, there is no need to manually update the app’s connection settings.
Usage details
Who can use it: Kimi Work desktop app versions 3.2.24 and later.
How to use it: Update Kimi Work to version 3.2.24 or later; the app will automatically use the operating system’s current proxy settings for network connections.
Tip: If proxy changes are frequent on your corporate network, keep the system setting as the single source of truth instead of configuring a separate fixed proxy in Kimi Work.
✓ Verified by official sourceSource: Kimi Work Release NotesKimi Work preview now supports Markdown note properties
With Kimi Work 3.2.24, Markdown files in the preview panel now support note properties. The editing and preview experience has also been improved.
Usage details
Who can use it: Kimi Work desktop app versions 3.2.24 and later.
How to use it: Open a Markdown file in a workspace in Kimi Work and switch to the preview panel; view and edit note properties fields there.
Tip: For structured notes, use note properties fields such as title, status, or tags to make files easier to scan.
✓ Verified by official sourceSource: Kimi Work Release NotesManaged Agent recovery prevents filling gaps without evidence
The Managed Agent recovery flow no longer fills in missing results without child ledger evidence, and more clearly distinguishes unknown and never-started states. This aims to reduce child tasks that incorrectly appear completed after recovery.
Usage details
Who can use it: Qwen Code Managed Agent recovery and failover flows.
How to use it: Reload a Managed Agent session after an interruption; check the recovery summary to confirm that items without child ledger evidence were not filled in and that states are differentiated.
Tip: Manually verify unknown states before delivering production output after automatic recovery.
✓ Verified by official sourceSource: Qwen Code v0.26.6 resmî sürüm notlarıManaged Agent upgrade uses the correct control plane endpoint
The qwen-managed-agent-upgrade start and verify steps now reach the control plane and Hosted Harness using the endpoint configuration resolved by the server. This prevents upgrades from targeting the wrong endpoint in deployments with different endpoint configurations.
Usage details
Who can use it: Operators managing Managed Agent deployments with qwen-managed-agent-upgrade.
How to use it: After loading the deployment configuration, run `qwen-managed-agent-upgrade start` and then `qwen-managed-agent-upgrade verify`.
Tip: If you use multiple endpoints, confirm the server’s resolved endpoint value in the logs before start.
✓ Verified by official sourceSource: Qwen Code v0.26.6 resmî sürüm notlarıManaged Session list shows a localized Retry notification on error
If the Managed Session list fails to load, Web Shell now displays a localized Retry notification and request details instead of an empty state. The error view appears below the Sessions and Archived tabs.
Usage details
Who can use it: Users of the Managed Sessions list in Web Shell.
How to use it: Open the Managed Sessions view in Web Shell; if the list fails to load, click Retry and, if needed, use the reported request ID for support investigation.
Tip: Check the Retry notification and request ID first to avoid mistaking a loading error for an empty list.
✓ Verified by official sourceSource: Qwen Code v0.26.6 resmî sürüm notlarıMessenger read and click notifications now go only to the sending app
The marketing_message_reads and marketing_message_clicks notifications are now delivered to the app that sent the message; other apps subscribed to the same Page do not receive them. This changes event ownership in integrations that use multiple apps.
Usage details
Who can use it: Messenger marketing message integrations that use multiple apps for the same Page.
How to use it: Update the relevant webhook subscription and event handler in the app that sent the message; reconfigure read and click logic in other apps so it does not rely on these notifications.
Tip: If you have multiple apps, centralize event ownership in the sending app and write events to a shared data store to prevent duplicate or missing records caused by missed notifications.
✓ Verified by official sourceSource: Messenger Platform değişiklik günlüğü — Okundu ve tıklama bildirimleri yalnızca gönderen uygulamaya gidermarketing_message_tracking_id added to Messenger click notifications
The Messenger marketing message click webhook now returns the marketing_message_tracking_id field. Matching this field to the same ID in the POST /act_<AD_ACCOUNT_ID>/messages response when sending the message lets you link the click to the send request.
Usage details
Who can use it: Developer apps that send Messenger marketing messages and use the marketing_message_clicks webhook.
How to use it: Read the marketing_message_tracking_id field in the marketing_message_clicks webhook payload; store the ID from the send response and match it to the click event. No separate setting is specified.
Tip: Store the ID in the send log alongside campaign and recipient context so you can reliably join click reports.
✓ Verified by official sourceSource: Messenger Platform değişiklik günlüğü — Tıklama bildirimlerinde izleme kimliğiMiMo Code passes the parent session’s reasoning variant to subagents
Subagents created with the actor tool now carry the reasoning variant selected for the parent turn. The variant is preserved for the same model or another model that offers the same level; the subagent’s own configured variant and an explicit --model selection take precedence.
Usage details
Who can use it: MiMo Code’s XiaomiMiMo/MiMo-Code main branch and agent flows that use actor. The change is available on the November 2, 2026 main branch, but no separate release tag has been published since November 1.
How to use it: Select a reasoning variant in the parent session and run a subagent with actor. If you change the subagent’s model, choose one that supports the same level name; if you give the subagent its own model or variant setting, or a --model override, that setting is used.
Tip: For long tasks, choose a level supported by both the parent and subagent; this prevents the subagent from falling back to its default reasoning setting.
✓ Verified by official sourceModels: mimo-v2.6-flashSource: MiMo Code resmi PR #2768: alt ajan varyant aktarımıSource: MiMo Code reasoning variants commit diffiMiMo Code uses model-supported reasoning variants and level names
With the official PR dated November 2, reasoning_options in models.dev now determines provider variants; the effort levels declared by the model are presented as-is, and protocol parameters no longer need to be entered manually for variants in config. Invalid low/medium/high options are removed for toggle-only models, while the existing behavior for budget-only and undeclared models is preserved.
Usage details
Who can use it: MiMo Code’s XiaomiMiMo/MiMo-Code main branch; providers using models.dev and models defined in config. The change is available on the November 2, 2026 main branch, but no separate release tag has been published since November 1.
How to use it: After updating from the main branch, use the level name declared in the model catalog in the variants field in config for the relevant model or agent; do not manually add the reasoning_effort protocol parameter. Do not show an effort level for a toggle-only model.
Tip: Check the levels supported by the model in its models.dev entry to avoid sending an unsupported reasoning_effort.
✓ Verified by official sourceModels: mimo-v2.6-flashSource: MiMo Code resmi PR #2768: reasoning variantsSource: MiMo Code reasoning variants commit diffiOpenAI increases Security Bug Bounty rewards and bonuses
OpenAI increased the midpoint of the reward ranges for medium, high, and critical findings in its Security Bug Bounty program by an average of more than fourfold. The base reward for a critical vulnerability in ChatGPT, Codex, or API Platform is $30,000, while the base reward for exceptional critical findings can reach $100,000; a high-quality report, priority campaign, and five-tier loyalty bonuses can bring the total to $172,000, and eligible previous reports are being reassessed retroactively.
Usage details
Who can use it: Independent security researchers participating in the program; scope and supported configurations are listed on the OpenAI Security Bug Bounty page on Bugcrowd.
How to use it: Read the scope and rules on OpenAI’s Bugcrowd Security Bug Bounty page; submit a report with a reproducible PoC, impact description, and steps using a currently supported configuration.
Tip: For a high-quality report, clearly isolate the impact and include a working PoC and precise reproduction steps; this increases the chances of qualifying for the quality multiplier.
✓ Verified by official sourceSource: OpenAI’nin daha yüksek bug bounty ödülleri duyurusuSource: OpenAI Security Bug Bounty programı koşullarıStartup, MCP, tool, and recovery speedups arrive for Qoder Fast
Startup via bytecode, asynchronous preparation, Bash/Git warm-up, direct return of suitable WebFetch output, on-demand loading of MCP definitions, safe parallel tool execution, and interruption recovery have been improved in the Qoder Fast execution layer. Use of some optimizations depends on the entry point and configuration.
Usage details
Who can use it: Desktop clients, Agent SDK, and custom applications that use Qoder CLI’s shared session runner; some options depend on the entry point/configuration.
How to use it: Select Fast from the model or tier selector in Qoder; for integrations using the CLI or SDK, start the task with the same session runner.
Tip: If many MCP tools are connected, prefer loading tool definitions on demand; starting a new session for unrelated work reduces context.
✓ Verified by official sourceSource: Qoder Fast modunu hızlandıran harness güncellemeleriQoder can resume stuck conversations after a background process crashes
Qoder 0.5.2 fixes an issue where a conversation would get stuck and could not continue when a background execution process failed. The goal is to make the conversation flow reusable after a failed process.
Usage details
Who can use it: Qoder 0.5.2 and later; conversations that run tasks in the background.
How to use it: Update the client to 0.5.2 or later; if a background process fails, reopen the same conversation and try the resume action again.
Tip: For long-running tasks, check the latest tool output before resuming after the update; don’t automatically repeat commands with external side effects.
✓ Verified by official sourceSource: Qoder 0.5.2 resmi sürüm notlarıQwen Code review: Text-based re-review requests are routed to the PR
The review flow routes re-review requests phrased in natural language to the correct pull request. This prevents prose re-review requests from being sent to the wrong target.
Usage details
Who can use it: Nightly releases v0.26.5-nightly.20261101.7b4b353750 and later; Qwen Code review flow.
How to use it: Request a re-review in natural language during a review session; the current nightly routes the request to the relevant PR context.
Tip: Clearly specify the PR context and the expected checks in the same sentence when requesting a re-review.
✓ Verified by official sourceSource: Qwen Code v0.26.5 nightly sürüm notlarıQwen Code Web Shell: Conversation Activity view added
Web Shell now lets you inspect a session’s conversation activity in a separate view. This makes it possible to track the session flow separately from the main workspace.
Usage details
Who can use it: Nightly versions v0.26.5-nightly.20261101.7b4b353750 and later; Qwen Code Web Shell.
How to use it: Upgrade to this nightly version and open the relevant session’s Conversation Activity view in Web Shell; the official note does not specify the menu path.
Tip: In long sessions, use the activity view to track status without interrupting the main conversation.
✓ Verified by official sourceSource: Qwen Code v0.26.5 nightly sürüm notlarıQwen Code Web Shell: Add to Split View added to the session menu
The Add to Split View action was added to session menus. Adding a session to the current split view no longer requires a separate flow.
Usage details
Who can use it: Nightly versions v0.26.5-nightly.20261101.7b4b353750 and later; Web Shell Split View.
How to use it: Open a session’s menu and select Add to Split View.
Tip: When comparing the context of two sessions, keep the main session fixed and add the auxiliary session to the split view.
✓ Verified by official sourceSource: Qwen Code v0.26.5 nightly sürüm notlarıQwen Code: 2>&1 redirection is no longer flagged as a file write
The shell permission check does not treat 2>&1, which only combines file descriptors, as a file write. This reduces unnecessary file-write confirmation prompts.
Usage details
Who can use it: Nightly releases v0.26.5-nightly.20261101.7b4b353750 and later; Qwen Code sessions that run shell commands.
How to use it: Use standard 2>&1 error stream redirection in a shell command; the new permission check will not classify it as a file write on its own.
Tip: Make permission prompts easier to review by avoiding mixing output redirection with actual file writes in the same command.
✓ Verified by official sourceSource: Qwen Code v0.26.5 nightly sürüm notlarıQwen Code: Unique MCP tools are preserved when tool names collide
When MCP tool names collide, tools with unique names are no longer dropped because of the collision. This makes tool discovery more reliable when using multiple MCP servers in the same session.
Usage details
Who can use it: Nightly versions v0.26.5-nightly.20261101.7b4b353750 and later; Qwen Code sessions using MCP servers.
How to use it: Enable multiple MCP servers in the same session using your usual connection flow; tools with unique names among the colliding names are preserved automatically.
Tip: Using unique prefixes in server and tool names reduces collisions and makes the tool list easier to understand.
✓ Verified by official sourceSource: Qwen Code v0.26.5 nightly sürüm notlarıQwen Code: 1 million-token context retained for Kimi K3
A nightly core fix preserves the 1 million-token context window throughout Kimi K3 generation. It aims to prevent context from unexpectedly shrinking during generation with long inputs.
Usage details
Who can use it: Nightly versions v0.26.5-nightly.20261101.7b4b353750 and later; Qwen Code sessions using Kimi K3.
How to use it: Connect the Kimi K3 model to nightly Qwen Code and run a long session as usual; no additional setting is required for this fix.
Tip: Avoid adding unnecessary files and old conversations to the context so you don’t consume the 1 million-token window too quickly.
Token Usage panel shows success rates and colors by model
The Web Shell Token Usage panel now includes the request success rate, number of failed requests, output speed, and model-level status indicators. When multiple models are used, it also shows a colored share bar and consistent model colors.
Usage details
Who can use it: All sessions using the Token Usage panel in Qwen Code Web Shell; it is also backward-compatible with older daemon versions.
How to use it: Open a session in Web Shell, open the Session token usage panel, and view the Overview or By model section.
Tip: When interpreting the success rate, keep in mind that retries are counted as separate requests.
✓ Verified by official sourceSource: Qwen Code v0.26.6 resmî sürüm notlarıSource: Token Usage başarı oranı ve renkleri değişikliğinin ayrıntılarıTypeScript SDK v0.1.27 bundles CLI 0.26.6
Qwen Code TypeScript SDK v0.1.27 has been released and bundles Qwen Code CLI 0.26.6. Projects using the SDK can get the current runtime with this release without installing the CLI separately.
Usage details
Who can use it: TypeScript/JavaScript projects using Node.js 22 or later; @qwen-code/sdk on npm.
How to use it: Run npm install @qwen-code/sdk@0.1.27 in your project and import the SDK from the @qwen-code/sdk package.
Tip: Pin the version for reproducible builds and verify the bundled CLI version in CI after upgrading the package.
✓ Verified by official sourceSource: Qwen Code TypeScript SDK v0.1.27 sürümüSource: @qwen-code/sdk kurulum ve API belgesiEmbedded runtime is automatically stopped before Windows updates
The Windows installer now stops the embedded runtime before replacing files. This reduces the likelihood of an update failing because node.exe is locked.
Usage details
Who can use it: Users updating with the Qwen Code Desktop Windows installer.
How to use it: Start the update from Desktop; the new installer automatically stops the runtime, replaces the files, and then restarts the app.
Tip: If a separate Qwen Code runtime is running during the update, close it too; this reduces the chance of remaining file locks.
✓ Verified by official sourceSource: Qwen Code v0.26.6 resmî sürüm notlarıBuilt-in review skill preserves model pairing in incremental reviews
During marker recovery, the built-in review skill requires the fetch-pr command to pass --since-model along with --since. The instruction also uses the deltaFiles and interaction[] scope instead of the now-unused since..head range.
Usage details
Who can use it: Qwen Code’s built-in review skill and incremental PR review workflow.
How to use it: For an incremental review, run the fetch-pr command in the form --since <sha> --since-model <model>; get both values from the same marker or review cache record.
Tip: Don’t infer the model identity from a separate source than the SHA; an incorrect pairing can cause the review to fall back to the full diff scope.
✓ Verified by official sourceSource: Qwen Code review skill anchor düzeltmesiSource: Qwen Code v0.26.6 sürüm notlarıACP client requests are queued while Kimi Code starts up
Kimi Code 2.5.0-canary now holds ACP initialize and other requests received before the server is ready until startup completes. This means clients such as Zed receive the normal startup flow or the actual startup error instead of the spurious Internal error seen during early handshakes.
Usage details
Who can use it: Zed and other ACP clients connecting to the Kimi Code ACP server; 2.5.0-canary preview.
How to use it: Upgrade to Kimi Code 2.5.0-canary.20261102033430-40590f6 and start the kimi acp command as usual; there’s no need to add a delay, even if the client sends its initialize request immediately after connecting.
Tip: Before removing artificial waits or retries from your custom ACP client, log the startup error body; if the service fails to start, the original provider startup reason is now passed through.
Background shell notifications are linked to the turn that created them
Background shell notifications are now associated with the turn that created them. This prevents notifications from being linked to the wrong message when multiple turns run concurrently.
Usage details
Who can use it: Qwen Code sessions that use background shell commands and concurrent turns.
How to use it: Run a shell command in the background and track its completion notification in the corresponding turn.
Tip: When using multiple background commands, check the turn link in the notification header to avoid acting on the result of the wrong command.
✓ Verified by official sourceSource: Qwen Code v0.26.6 resmî sürüm notlarıArchived sessions are hidden from the session list immediately
Kimi Code 2.5.0-canary fixes old session rows reappearing in the list during the brief interval between archiving and the session index being updated on disk. It also prevents stale or duplicate sessions from appearing in paginated lists.
Usage details
Who can use it: Kimi Code CLI, kimi web, and Desktop/Web session lists; 2.5.0-canary preview.
How to use it: Upgrade Kimi Code to 2.5.0-canary.20261102033430-40590f6; after archiving a session, leave it in the normal session list—it should no longer appear, even before the index is written to disk.
Tip: Use the archive filter to view archived conversations. Since brief inconsistencies between the normal list and archive list are no longer expected, automated listers should need less additional deduplication.
Capture cards use DejaVu fonts in the sandbox
Capture cards in the sandbox environment now load DejaVu fonts for the intended monospace appearance. This makes lines in code and output cards easier to read.
Usage details
Who can use it: Qwen Code Web Shell workflows that generate sandbox capture cards.
How to use it: Create a capture card in the sandbox environment and check the monospace alignment of code blocks.
Tip: Use the same version of the sandbox image to rule out font differences when comparing visual output.
✓ Verified by official sourceSource: Qwen Code v0.26.6 resmî sürüm notlarıRedirections in cd/pushd segments are now reported
The CLI now reports redirections that occur within `cd` command segments. This makes the effects of shell redirection visible alongside changes to the working directory.
Usage details
Who can use it: All sessions using Qwen Code CLI shell commands.
How to use it: Run a shell command containing `cd` or `pushd` in a session and inspect the redirect report in the command notification.
Tip: If you are not expecting a file-writing redirection, use this new report to investigate unexpected changes to permissions or security rules.
✓ Verified by official sourceSource: Qwen Code v0.26.6 resmî sürüm notlarıUnsupported Mod hooks no longer stop startup
If a Mod manifest contains unsupported hooks such as prompt.paint or ui.place, Qwen Code now skips them without stopping startup. Users are warned about which hooks are unsupported.
Usage details
Who can use it: CLI, Web Shell, or related hosts using Qwen Code Mods; hooks the host does not support.
How to use it: Load the Mod as usual; check the startup logs for warnings about unsupported hooks and rely only on hooks supported by the host.
Tip: If a Mod’s critical behavior depends on an unsupported hook, monitor startup warnings in CI as failure signals to verify that it is not being silently skipped.
✓ Verified by official sourceSource: Qwen Code v0.26.6 resmî sürüm notlarıTypos in disabled MCP tool names now trigger warnings on every discovery
MCP discovery flows now warn on every discovery path when disabled tool names contain typos. This makes it clear why a tool is missing not just on the first discovery, but on subsequent attempts as well.
Usage details
Who can use it: Qwen Code users with MCP servers and disabled tool configuration.
How to use it: Add a tool to the disabled list in the MCP configuration, run discovery, and compare the tool name in the warning with your configuration.
Tip: Copy tool names from the server’s actual discovery output; manually typed names are the most common cause of this warning.
✓ Verified by official sourceSource: Qwen Code v0.26.6 resmî sürüm notlarıLoop-protection reminder clears only when the request is accepted
The Loop-protection reminder and related telemetry are now considered complete only when the model request carrying the reminder is actually accepted. Protection state is not cleared prematurely for failed or rejected requests.
Usage details
Who can use it: Qwen Code sessions using Loop-protection.
How to use it: Create a recurring tool loop and, when the reminder appears, use the session logs to check whether the model request was accepted.
Tip: If the request was rejected, assume protection is still active in the next turn; do not treat it as having been cleared early.
✓ Verified by official sourceSource: Qwen Code v0.26.6 resmî sürüm notlarıSuccess messages no longer appear before file downloads are complete
Kimi Work 3.2.24 fixed an issue that caused a success message to appear before a file had finished downloading. This better aligns the success notification with the file being ready to use.
Usage details
Who can use it: Kimi Work desktop app versions 3.2.24 and later.
How to use it: Update Kimi Work; when downloading a file or artifact, check that the file is complete after the success notification before opening it.
Tip: For large files, verify the file’s size and that it opens instead of moving it as soon as the success notification appears.
✓ Verified by official sourceSource: Kimi Work Release NotesAI attribution notes use the correct diff for git commit --amend
For loosely written `git commit --amend` commands, AI attribution notes now generate a diff for the commit actually being amended instead of previous commits. This prevents the attribution record from being linked to the wrong set of changes.
Usage details
Who can use it: Qwen Code CLI users who enable AI attribution notes.
How to use it: With attribution enabled, update a commit using `git commit --amend` or an equivalent loosely written command, then check that the generated note uses the diff for the amended commit.
Tip: After amending, compare the attribution note with the output of `git show --stat HEAD`.
✓ Verified by official sourceSource: Qwen Code v0.26.6 resmî sürüm notlarıKimi Work Network Diagnostics screen can now open offline
Kimi Work 3.2.24 allows the network diagnostics screen to open even without an internet connection. This makes it easier to access the diagnostic tool when connection problems occur.
Usage details
Who can use it: Kimi Work desktop app versions 3.2.24 and later.
How to use it: Open the Network Diagnostics screen in Kimi Work; you can view the screen and inspect the current network status even when there is no connection.
Tip: During corporate network outages, take a screenshot or capture the output from the diagnostics screen before sending it to the support team.
✓ Verified by official sourceSource: Kimi Work Release NotesKimi Work Remote Control connection stability improved
Kimi Work 3.2.24 improved the stability of Remote Control connections. Connection drops and reconnection behavior have been improved in sessions that use remote control.
Usage details
Who can use it: Kimi Work desktop app versions 3.2.24 and later; desktop and remote clients using Remote Control.
How to use it: Update Kimi Work to version 3.2.24 or later and restart the Remote Control flow using the existing connection/QR connection.
Tip: For long-running remote tasks, preventing the desktop device from going to sleep helps you make better use of the improved connection stability.
✓ Verified by official sourceSource: Kimi Work Release NotesKimi Work system notification toggle issue fixed
Kimi Work 3.2.24 fixed an issue where the system notifications toggle did not respond when clicked. The setting now responds correctly to user input.
Usage details
Who can use it: Kimi Work desktop app versions 3.2.24 and later.
How to use it: Update the app to version 3.2.24 or later, then use the system notifications toggle in notification settings again.
Tip: If you are not receiving notifications, check both the Kimi Work toggle and the operating system’s app notification permission.
✓ Verified by official sourceSource: Kimi Work Release NotesLinux AppImage packages check for required dynamic libraries
Linux AppImage builds now verify that required dynamic libraries are either included in the package or must be provided by the host. This makes missing libraries that prevent the app from launching more visible before installation.
Usage details
Who can use it: Users of Qwen Code Desktop Linux AppImage.
How to use it: Download the new AppImage from GitHub Releases, run it, and if a missing-library notice appears at launch, install the libraries using your distribution’s package manager.
Tip: Test launch on a clean Fedora/RHEL/Ubuntu image before deploying the AppImage to a production machine.
✓ Verified by official sourceSource: Qwen Code v0.26.6 resmî sürüm notlarıManaged Agent no longer leaves consumed child results pending
The delivery status of consumed child results in Managed Agent no longer remains pending. This prevents results that have already been received from appearing as still pending in session state and the recovery screen.
Usage details
Who can use it: Sessions using Managed Agent child tasks and result delivery.
How to use it: In a Managed Agent session, consume child task results and refresh the session state; check that delivered results are no longer pending.
Tip: After recovery or reconnection, reprocess only records that remain pending.
✓ Verified by official sourceSource: Qwen Code v0.26.6 resmî sürüm notlarıMiMo Code SDK adds efforts and reasoningBudget fields to Model.capabilities
Optional efforts and reasoningBudget fields have been added to the MiMo Code SDK’s Model.capabilities type. Clients can now read the effort levels supported by a model and its budget-based reasoning support from the provider catalog, reducing the need to maintain a separate provider parameter table.
Usage details
Who can use it: MiMo Code packages/sdk main branch; the change appears in the official PR dated November 2, 2026. No new release tag has been published since November 1 for the packaged version.
How to use it: After updating the SDK from the main branch, read model.capabilities?.efforts and model.capabilities?.reasoningBudget on the Model object to build interface or request options; if efforts is empty, do not show a level selector.
Tip: An undefined efforts field means no catalog declaration is available, while an empty array means the model offers no effort levels; do not interpret these two cases the same way.
✓ Verified by official sourceSource: MiMo Code resmi PR #2768: SDK alanlarıSource: MiMo Code reasoning variants commit diffiMobile Web Shell composer layout adapts to narrow screens
On narrow screens, the Web Shell composer now aligns workspace controls, truncates long names, and centers the context-usage ring. This reduces input-field overflow and control clutter in the mobile view.
Usage details
Who can use it: People using Web Shell on a mobile device or in a narrow browser window.
How to use it: Open Web Shell on a mobile device or at a narrow window width; check the workspace name, controls, and context-usage ring around the composer.
Tip: Switch to a wider screen view when you need to see the full name to distinguish long workspace names.
✓ Verified by official sourceSource: Qwen Code v0.26.6 resmî sürüm notlarıQoder fixes startup failure caused by invalid execution environment data
Qoder 0.5.2 fixes an issue where invalid execution environment data in some older conversations caused startup and conversation-loading failures. Such historical records should no longer break app startup or conversation loading in the same way.
Usage details
Who can use it: Qoder 0.5.2 and later; users loading older conversation records.
How to use it: Update Qoder to 0.5.2 or later, then reload the conversation that previously could not be opened.
Tip: If the issue occurs only with older conversations, update the client before starting a new task; try reloading the records instead of deleting them.
✓ Verified by official sourceSource: Qoder 0.5.2 resmi sürüm notlarıA malformed task record no longer prevents other tasks from loading in Qoder
Qoder 0.5.2 fixes an issue where malformed task records prevented other valid records in the task list from loading. The entire list should no longer become unusable because of a single corrupted record.
Usage details
Who can use it: Qoder 0.5.2 and later; all platforms that use the task list.
How to use it: Upgrade the Qoder client to 0.5.2 or later, then reopen or refresh the task list.
Tip: If refreshing the list doesn’t resolve the issue, open valid tasks first and back up critical outputs; consider removing the old record only as a last resort.
✓ Verified by official sourceSource: Qoder 0.5.2 resmi sürüm notlarıQwen Code CLI: PR badge is parsed correctly when using an adjacent --flag
The PR badge is no longer parsed incorrectly when a --flag is written adjacent to the PR number in the review flow. This preserves the correct PR context in CLI calls that use adjacent flags.
Usage details
Who can use it: Nightly versions v0.26.5-nightly.20261101.7b4b353750 and later; Qwen Code review CLI flow.
How to use it: Run the review command with the PR number and the --flag you use; the nightly version will associate the PR badge correctly.
Tip: For more readable command-line parsing, prefer putting a space between the PR number and the flag.
✓ Verified by official sourceSource: Qwen Code v0.26.5 nightly sürüm notlarıQwen Code CLI: Long /btw responses are readable in a constrained viewport
Long /btw responses in the CLI now remain readable in a constrained viewport. Responses are less likely to be cut off in a narrow or fixed-height view.
Usage details
Who can use it: Nightly versions v0.26.5-nightly.20261101.7b4b353750 and later; Qwen Code CLI.
How to use it: Run the /btw command in the CLI; long responses will be displayed using the new constrained-viewport behavior.
Tip: Use /btw to ask brief follow-up questions without interrupting the main task, and read the long response in the viewport.
✓ Verified by official sourceSource: Qwen Code v0.26.5 nightly sürüm notlarıQwen Code Web Shell: Back control outside the nav rail for the Plugins page
A back control is now visible for returning from the Plugins page when the navigation rail is not shown. This keeps plugin navigation usable in collapsed layouts or layouts without the rail.
Usage details
Who can use it: Nightly versions v0.26.5-nightly.20261101.7b4b353750 and later; Web Shell Plugins page.
How to use it: With the nav rail turned off in Web Shell, open the Plugins page and use the back control on the page.
Tip: When leaving plugin details, prefer the page’s own back control over the browser’s back button.
✓ Verified by official sourceSource: Qwen Code v0.26.5 nightly sürüm notlarıQwen Code: Clearer query rejection when no LSP server is ready
LSP queries are no longer sent when no server is ready; they are explicitly rejected instead. This provides a clearer error flow indicating that the server is not ready, rather than a silent or ambiguous result.
Usage details
Who can use it: Nightly versions v0.26.5-nightly.20261101.7b4b353750 and later; Qwen Code projects with LSP enabled.
How to use it: In a project with LSP enabled, wait for the server to start before running the query; if the server is not ready, look for the new explicit error message.
Tip: Before running an LSP query, confirm that the relevant language server is enabled in the project and workspace settings.
✓ Verified by official sourceSource: Qwen Code v0.26.5 nightly sürüm notlarıThe Hunger Games: Sunrise on the Reaping is coming to Meta VR Glasses in 3D
Meta has partnered with Lionsgate to convert The Hunger Games: Sunrise on the Reaping into 3D. The film will be released in 3D theaters on November 20, 2026, and will also be available to watch in 3D on Meta VR Glasses starting in spring 2027.
Usage details
Who can use it: The 3D version for Meta VR Glasses is planned for spring 2027; the theatrical 3D release was announced for November 20, 2026.
How to use it: No device settings are needed for now; buy a 3D ticket for the theatrical release. When it becomes available on Meta VR Glasses in spring 2027, check the film library for the 3D version of The Hunger Games: Sunrise on the Reaping.
Tip: If you’re aiming for the 3D experience, plan your theater visit early; for the Meta VR Glasses version, check the library and purchase options separately at launch.
✓ Verified by official sourceSource: Meta’nın Lionsgate ile The Hunger Games: Sunrise on the Reaping 3D ortaklığı duyurusuverify report separates targeted gate results from round assertions
The verify report now counts the results of the targeted gate test suite separately from the number of scripted assertions in the round. This ensures the assertion total reflects only the checks for that round.
Usage details
Who can use it: Qwen Code verify-pr reports and published verification comments.
How to use it: Run a `/verify-pr` round and review the generated report’s separate fields for round assertions and targeted gate test-suite results.
Tip: When assessing a round’s success, look not only at the total but also at the separate section for targeted gate results.
✓ Verified by official sourceSource: Qwen Code v0.26.6 resmî sürüm notlarıverify-pr correctly reports closed PR skips and suite counts
Verify-pr counts each test suite run as one assertion and correctly handles skip notices on closed PRs. This prevents a verification round from being mistakenly rejected for having zero assertions.
Usage details
Who can use it: GitHub PR workflows using Qwen Code verify-pr, especially closed PRs and suite-level checks.
How to use it: Run `/verify-pr` or the existing monitoring workflow on a PR; check that the report reflects one assertion per suite and the closed-PR skip notice results.
Tip: When you want to review a closed PR again, treat the skip notice in the report as scope information, not as an error.
✓ Verified by official sourceSource: Qwen Code v0.26.6 resmî sürüm notlarıVS Code session history now uses calendar weeks instead of a rolling 7 days
Session history in Qwen Code Companion’s VS Code webview is now grouped by calendar week starting on Monday instead of a rolling 168 hours. As a result, the previous Saturday appears under Older rather than This Week at the start of the week.
Usage details
Who can use it: Qwen Code Companion’s VS Code webview session history.
How to use it: Open the Qwen Code Companion chat view in VS Code and check the This Week and Older groups in the session history dropdown.
Tip: On Mondays, look for sessions from the previous week under Older instead of This Week; search and the view layout are unchanged.
✓ Verified by official sourceSource: Qwen Code için VS Code takvim haftası düzeltmesiSource: Qwen Code v0.26.6 sürüm notlarıWorkflow mode is more visible in the composer + menu
The Workflow mode toggle has moved from the composer toolbar to the composer’s + menu. Web Shell now shows the active mode as a dismissible chip, a warning when tools are unavailable, and a brief description in the + menu.
Usage details
Who can use it: Users whose sessions use the Workflow mode setting in Web Shell.
How to use it: Open the + menu in the composer, then select Workflow mode; if it is active, turn it off using the displayed Workflow chip.
Tip: In sessions where tools are unavailable, heed the warning on the chip and return to normal mode instead of forcing Workflow.
✓ Verified by official sourceSource: Qwen Code v0.26.6 resmî sürüm notları/autofix-pr allows up to three pushes for failing checks
If the same check is still failing after three fix pushes, `autofix-pr` marks it as requiring human intervention instead of sending it back to the agent. This entry appears in `/autofix-pr status`, while watch continues monitoring the other checks.
Usage details
Who can use it: Qwen Code PR watch/autofix-pr; up to three failed fix pushes per check.
How to use it: Run `/autofix-pr status`; fix the issue manually for a check awaiting human intervention, or reset the entry with a new successful head.
Tip: When a check fails for the third time, don’t repeat the same fix approach; review the logs and CI configuration as a human.
✓ Verified by official sourceSource: Qwen Code autofix retry-budget PRSource: Qwen Code v0.26.5 release notesA shared-work mode using --in-place has been added for /autofix-pr
In a clean checkout on the PR head branch, `/autofix-pr <n> --in-place` now runs the agent in the foreground in the user’s own working tree, with an approval flow. At the end of the run, the agent pushes the branch and responds to the PR using the user’s identity; it stops on a dirty or diverged checkout and asks the user to decide what to do.
Usage details
Who can use it: Qwen Code CLI; a clean checkout on or behind the PR head that tracks a named remote; single-run, supervised use.
How to use it: Start a session in the root checkout of the PR head branch and run `/autofix-pr <PR-number> --in-place`; `--every`, scheduled, and headless use are not supported in this mode.
Tip: Commit or stash your own changes before the agent starts; this mode explicitly notes that it can also push the user’s commits.
✓ Verified by official sourceSource: Qwen Code --in-place co-driving PRSource: Qwen Code v0.26.5 release notesChatGPT Work’s agent capabilities are coming to the main ChatGPT experience
OpenAI is preparing to bring Work’s agent capabilities to the standard Chat experience. According to the November 1 announcement, the combined Chat and Work experience will be tested with a small group of users in the coming weeks; selected users will be able to get work done on the web with files and plugins without switching modes. OpenAI’s help page confirms that these capabilities will come to the main ChatGPT experience in December.
Usage details
Who can use it: A small test group in the coming weeks; broader rollout in December 2026. The announcement of a shared quota covers Plus, Pro and Business plans; Codex remains a separate experience in the ChatGPT desktop app.
How to use it: If you’re included in the test, start a conversation in the regular Chat tab; add a file or connected plugin, then give ChatGPT a task directly, such as building a website or preparing a presentation, and review the work. There’s no need to switch to a separate Work mode.
Tip: After brief idea research, clearly assign the target task in the same conversation; check file and plugin permissions in advance to reduce unnecessary pauses by the agent.
CLI update can target an exact version without npm discovery
On standalone installations, `qwen update --target-version <version>` downloads, verifies, and installs a specific version without discovering releases through npm. A `v` prefix, reinstalling the same version, and downgrades are supported; the pinned version preference is not persistent.
Usage details
Who can use it: Standalone Qwen Code CLI installations; access to the selected release archive is required, including in environments without npm registry access.
How to use it: Run `qwen update --target-version 0.26.5`; set `QWEN_UPDATE_BASE_URL` if you need a custom archive root. For a persistent older version, set automatic updates to `general.enableAutoUpdate: false`.
Tip: For production deployments, pin the target version only after testing checksum verification and the rollback archive.
✓ Verified by official sourceSource: Qwen Code exact-version update PRSource: Qwen Code v0.26.5 release notesInteractive card values are carried into the next message as context
After a value change in a card remains stable for about 800 ms, it is retained as a removable tag above the Web Shell composer, and the next question receives the current values. The built-in card’s “Continue analysis with current parameters” action now sends the tagged reference directly without opening a review window.
Usage details
Who can use it: Web Shell sessions using Generated UI; only when the card is visible, active, and linked to the same session.
How to use it: Change a card value, wait for the tag to appear, and then ask your question; remove the tag above the composer if you don’t want it. To continue from the card, click “Continue analysis with current parameters.”
Tip: Keep the tag to carry the card’s latest parameters into a calculation question instead of re-entering the individual values.
✓ Verified by official sourceSource: Generated UI chat continuity and default-on PRSource: Qwen Code v0.26.5 release notesGenerated UI enabled by default in the stable release
With Qwen Code v0.26.5, Generated UI is now enabled by default in Web Shell without requiring configuration. This expands access from the previous nightly default to the stable release; MCP Apps views aren’t available in terminal, headless, or channel sessions.
Usage details
Who can use it: Qwen Code v0.26.5 Web Shell and Desktop; supported trusted workspaces. Can be disabled with `experimental.generatedUi: false`.
How to use it: Open Web Shell in a new session without adding any settings; to restore the previous behavior, set `experimental.generatedUi` to `false` in settings.json and restart the session.
Tip: If a view can’t be created in untrusted workspaces, use the text fallback; the default behavior doesn’t generate unnecessary warnings in this case.
✓ Verified by official sourceSource: Qwen Code v0.26.5 release notesSource: Generated UI chat continuity and default-on PRCost-related restriction reported on Grok Bot’s weekly usage quota
On November 1, some Grok Bot users reported that their weekly usage percentage quickly reached 100% despite not using the service; TechCrunch and The Verge reported this as a restriction. Elon Musk confirmed that Grok Bot’s service costs were unsustainably high relative to revenue but did not disclose new quota figures or how much each plan was affected.
Usage details
Who can use it: Reported to affect the shared weekly usage pool on paid Grok/X Subscription plans; new limits by plan and country availability were not disclosed.
How to use it: Open Settings → Usage in Grok Bot and check the remaining percentage, usage breakdown, and reset time; if needed, review top-up or higher-plan options.
Tip: Spread heavy tasks across the week and track usage on this screen; Grok’s official account says heavy tasks consume the pool faster.
Experimental one-command Helm installation arrives for Managed Agent
The Managed Agent control plane gains an experimental Helm installation with a Hosted Harness and persistent MySQL working together. The package also includes optional trusted single-node LOCAL operation and an authenticated gateway that exposes only the Managed interface.
Usage details
Who can use it: Experimental Kubernetes deployment; signed Managed Agent, Hosted Harness, MySQL, and authenticated gateway components.
How to use it: Choose the experimental Helm path in the Qwen Code Managed Agent deployment documentation; provide image, model configuration, and persistent MySQL values, and set the gateway credentials.
Tip: Before deploying to production, test that the gateway rejects ordinary daemon paths and verify credential and database protection after uninstalling.
✓ Verified by official sourceSource: Qwen Code experimental Kubernetes deployment PRSource: Qwen Code v0.26.5 release notesManaged WebShell now paginates upward through live messages
Managed WebShell now reads the latest 100 live items directly and loads older messages by paginating upward. Live updates continue during pagination.
Usage details
Who can use it: Managed WebShell sessions; the latest 100 live items in the initial view, with upward scrolling for older history.
How to use it: Open a Managed WebShell session and scroll up in the message list; older records are added as they load, while new messages appear in the live stream.
Tip: To quickly reach a section in a long history, scroll up first to load an older page, then return to the live stream.
✓ Verified by official sourceSource: Qwen Code v0.26.5 release notesSource: Qwen Code Desktop v0.26.5 release notesA shared ChatGPT–Codex token quota is coming to Plus, Pro and Business
Starting in December 2026, OpenAI will combine ChatGPT and Codex usage into a single shared quota for Plus, Pro and Business plans. Usage will be measured in tokens based on the selected model and task; Instant chat won’t count against this quota. If you reach the limit, you’ll be able to wait for the reset, upgrade your plan, or purchase credits for eligible features.
Usage details
Who can use it: Starting in December 2026; Plus, Pro and Business plans, as well as ChatGPT and Codex. Instant chat isn’t included in the shared usage quota.
How to use it: Open Settings → Usage in ChatGPT to track your remaining combined usage and reset time. If you reach the limit, wait for the reset, upgrade your plan, or purchase credits for eligible features.
Tip: Save the shared quota for long and complex tasks; use Instant chat for short everyday questions, since according to OpenAI, this usage doesn’t consume the quota.
✓ Verified by official sourceSource: ChatGPT ve Codex kullanım limitlerini anlama – OpenAI Yardım MerkeziSource: Kullanım kotası ChatGPT ve Codex arasında paylaşılacak – OpenAI Topluluk ForumuSource: OpenAI, Pro kullanım limitlerini ikiye katlayıp Work özelliklerini Chat’e getiriyor – BleepingComputerOpt-in group moderation tools added to QQ bot
Channel adapters can now expose their own tools to agent sessions; the QQ bot is the first to use this, adding 18 moderation tools, including recall, mute/unmute, kick, blacklist, group info, and join request review. The tools are enabled with the `moderationTools` key, which is off by default.
Usage details
Who can use it: Qwen Code QQ group channels; operations that require the bot to be a group admin and `moderationTools` to be explicitly enabled by the operator.
How to use it: Set `moderationTools: true` in the QQ channel configuration; run the bot as a group admin and call `qq_recall_message`, `qq_set_mute`, `qq_kick_member`, or the relevant tool in the channel session.
Tip: Test in a separate group before enabling destructive operations; enable the key only when needed to preserve the safe default.
✓ Verified by official sourceSource: Qwen Code channel tools and QQ moderation PRSource: Qwen Code v0.26.5 release notesSession Workflow switch added to Web Shell and VS Code
Web Shell, split pane, and modern VS Code chat can now change a saved session’s Workflow preference using a separate switch. The change takes effect in the next new round without interrupting the response.
Usage details
Who can use it: Supported Web Shell sessions, split pane, and modern VS Code chat; new chats show only an informational placeholder.
How to use it: Turn the Workflow switch in the composer toolbar of a supported session on or off; the preference takes effect in a new message round.
Tip: To avoid interrupting a long-running response, change the switch while the current response is in progress and expect it to take effect in the next round.
✓ Verified by official sourceSource: Qwen Code session Workflow controls PRSource: Qwen Code v0.26.5 release notes/verify-pr reuses a clean PR working tree
The local `/verify-pr` flow reuses the current working tree only if it is a clean checkout of the PR head. Details of verification rounds are also recorded.
Usage details
Who can use it: Local `/verify-pr` flow in the Qwen Code CLI; requires a clean checkout matching the PR head.
How to use it: Run `/verify-pr` from the root checkout of the PR head branch; if the branch is clean and matches the PR head, the existing tree is used; otherwise, the flow requires new preparation.
Tip: Clean the working tree with `git status` before verification; do not expect local changes to be reset automatically.
✓ Verified by official sourceSource: Qwen Code v0.26.5 release notes/verify-pr rounds keep reports and comments separate
Each `/verify` round now updates only its own status comment; the latest round is published, and the verdict line is validated as a plain paragraph. This prevents comments from previous rounds from being overwritten accidentally.
Usage details
Who can use it: Qwen Code GitHub `/verify-pr` publishing flow.
How to use it: Run multiple verification rounds and check each round’s separate status in the PR comments; use the latest report as the source of truth.
Tip: Compare rounds by their IDs instead of deleting old comments; use the latest published round to make a decision.
✓ Verified by official sourceSource: Qwen Code v0.26.5 release notesExport and restore actions are now visible directly on archived conversations
Archived conversations in Settings now show Export conversation and Restore actions inline, next to the More menu. This reduces the number of menu steps required to manage archives.
Usage details
Who can use it: Qwen Code Settings > Archived conversations.
How to use it: Open Settings, go to Archived conversations, and click Export conversation or Restore on the relevant row.
Tip: Export conversations you plan to keep for a long time before archiving them to simplify the list.
✓ Verified by official sourceSource: Qwen Code v0.26.5 release notesA blocked autofix run leaves an explanatory comment on the PR
When an `autofix-pr` run is blocked, it now posts a comment on the PR explaining what was attempted, what blocked it, and what human action is needed. This keeps pending work from being recorded only in a local state file.
Usage details
Who can use it: GitHub PRs using Qwen Code PR watch/autofix-pr.
How to use it: Check the `/autofix-pr` status or PR comments, and complete the human action specified in the comment created for the blocked run.
Tip: Resolve the stated blocker before closing the comment; start a new run with a new head instead of blindly rerunning the same one.
✓ Verified by official sourceSource: Qwen Code v0.26.5 release notesContext Usage separates instruction files from auto memory
The CLI’s context usage view now shows instruction files separately from automatic memory. This makes it clearer how much each type of file contributes to the context.
Usage details
Who can use it: Qwen Code CLI `/context` and the Context Usage view.
How to use it: Run `/context` during a session and review the instruction files and auto memory sections separately.
Tip: When approaching the context limit, reduce large instruction files first, then auto memory content.
✓ Verified by official sourceSource: Qwen Code v0.26.5 release notesMemory files can be previewed from the Context Usage panel
Context memory files in the Context Usage panel can now be clicked to open a read-only preview of files such as AGENTS.md. Users can directly inspect which files contribute to the context.
Usage details
Who can use it: Qwen Code CLI/Web Shell Context Usage panel; only memory files included in the context.
How to use it: Open the Context Usage panel, click a memory file row, and inspect the read-only preview that opens.
Tip: When context consumption rises, use this preview to identify large files first.
✓ Verified by official sourceSource: Qwen Code v0.26.5 release notesCUA SDK session and window operations aligned with v0.27.0
The CUA SDK, Browser SDK, and Node REPL in Qwen Code v0.26.5 align their login, window selection, and session cleanup behavior with CUA Driver v0.27.0. The goal is more reliable computer-use sessions.
Usage details
Who can use it: Qwen Code CUA integration, Browser SDK, and Node REPL; with CUA Driver v0.27.0.
How to use it: Upgrade Qwen Code to v0.26.5 and CUA Driver packages to v0.27.0; restart existing CUA sessions.
Tip: After upgrading CUA, verify the window selection and session-closing flow with a quick test on your operating system.
✓ Verified by official sourceSource: Qwen Code v0.26.5 release notesSource: Qwen Code releases list with CUA Driver v0.27.0One-click daemon updates support pnpm, yarn, and bun
Web Shell’s one-click daemon update now also covers pnpm, yarn, bun, and sudo-prefixed npm installations. If automatic updating isn’t possible, the interface shows the steps to apply manually.
Usage details
Who can use it: Web Shell daemon update flow; users whose installation method is pnpm, yarn, bun, or sudo npm.
How to use it: Open the daemon update prompt in Web Shell and select the one-click update; if the automatic route isn’t suitable, run the manual commands shown.
Tip: If the daemon is installed outside the system PATH, use the manual route suggested by the interface; don’t run a random npm global update.
✓ Verified by official sourceSource: Qwen Code v0.26.5 release notesThe status bar shows the number of running background agents
Web Shell and Qwen Code Desktop show the number of active agents in the task capsule in the status bar while background agents are running. This lets you see which sessions are still working without opening them.
Usage details
Who can use it: Qwen Code Web Shell and Desktop; sessions with running background agents.
How to use it: Start multiple background agents and watch the task capsule in the status bar; the count updates as the running agents change.
Tip: Don’t close the app until the count in the capsule reaches zero; use this indicator to quickly track long-running tasks.
✓ Verified by official sourceSource: Qwen Code v0.26.5 release notesKimi Design: Editable visual design surface made available to users
Moonshot’s official sources dated November 1 present Kimi Design as a user-facing surface that generates posters, social media graphics, product images, and infographics from text prompts or reference images, and offers element-level editing on a canvas, comment-based revisions, version history, and export. This is an accessibility listing for Kimi Design, which was not included in the known list.
Usage details
Who can use it: In the Kimi web interface; official sources do not specify plan or regional restrictions.
How to use it: Open https://www.kimi.com/design; write a prompt or upload a reference image, edit the generated design on the canvas or by commenting on an element, then export the result as PNG, JPG, PDF, PPTX, or Google Slides.
Tip: Specify the target audience, canvas size, colors, and output format in the initial prompt; make small revisions by commenting on individual elements.
Managed Agent upgrade command handles interrupted installations more safely
The check, backup, install, start, and verify stages of the Managed Agent upgrade command have been hardened to clean up and verify after interrupted installations. This reduces the risk of leftover temporary files and failed startup if an upgrade is interrupted.
Usage details
Who can use it: Operators upgrading the Qwen Code Managed Agent reference deployment.
How to use it: First run the upgrade command’s check stage; then complete the start and verify steps and check the temporary upgrade directory after an interruption.
Tip: Do not start a new upgrade without manually checking the temporary upgrade directory that may be left behind after SIGKILL.
✓ Verified by official sourceSource: Qwen Code v0.26.5 resmî sürüm notlarıNative extension commands require approval before installation
In OpenTUI, native extension commands now require user approval before being installed directly. Replacing or unmounting an extension cancels pending approval requests.
Usage details
Who can use it: Qwen Code’s OpenTUI interface and native extension commands.
How to use it: When a native extension command requests installation, review the approval prompt and explicitly accept it; if you replace or unmount the extension, restart the pending request.
Tip: Don’t approve native extension commands whose source you don’t recognize; this approval boundary prevents direct execution.
✓ Verified by official sourceSource: Qwen Code v0.26.5 release notesPR watch merges a stale base before attempting a fix
When a CI failure may be caused by the branch being behind its base, PR watch now merges the stale base before starting a fix. This reduces the risk of triggering an unnecessary fix cycle.
Usage details
Who can use it: Qwen Code PR watch and autofix-pr flows.
How to use it: Run PR watch as usual; if the branch is behind, the flow applies the stale base merge step before attempting a fix.
Tip: Ensuring the PR is synchronized with the base before an automatic fix helps reduce unrelated merge conflicts.
✓ Verified by official sourceSource: Qwen Code v0.26.5 release notesQwen Code Managed: Paginated SQL history queries added
Bounded SQL history queries have been added for Managed local H2 and Hosted MySQL. These queries support paginated retrieval of recent messages, journal transaction records, and specific operation or command searches.
Usage details
Who can use it: Qwen Code v0.26.5 and Qwen Code Desktop v0.26.5; installations using Managed local H2 or Hosted MySQL.
How to use it: Upgrade to Qwen Code v0.26.5 and use pagination or the operation/command lookup flow in the Managed history query API; the official note does not specify the exact parameter names.
Tip: Fetch recent messages in pages and narrow operation/command searches; avoid requesting large history tables in a single query.
✓ Verified by official sourceSource: Qwen Code v0.26.5 sürüm notlarıQwen Code: Bodyless 400 responses no longer fail immediately
Bodyless HTTP 400 responses are now retried instead of being treated as fail-fast. This makes tasks more resilient to transient or malformed network responses that have no response body.
Usage details
Who can use it: Qwen Code v0.26.5 and Qwen Code Desktop v0.26.5.
How to use it: Upgrade Qwen Code to v0.26.5; retries for bodyless 400 responses are applied automatically.
Tip: For intermittent request failures, wait for the built-in retry instead of restarting the task immediately.
✓ Verified by official sourceSource: Qwen Code v0.26.5 sürüm notlarıSandboxed /verify-pr checks artifacts before publishing
The publisher for `/verify-pr` inside the sandbox now runs the artifact check itself. If the check cannot run or fails, the verification round is marked inconclusive.
Usage details
Who can use it: Qwen Code sandboxed `/verify-pr` publishing flow.
How to use it: Run `/verify-pr` and inspect the artifact check result in the published report; if the round is inconclusive, fix the missing artifact and start a new round.
Tip: Do not treat the verification result as successful until the artifact check passes.
✓ Verified by official sourceSource: Qwen Code v0.26.5 release notesBulk live session snapshots for 20 workspaces arrive in the SDK
A new batch request and SDK method retrieve the live session status of up to 20 selected workspaces in a single call. This reduces the need for separate requests in multi-workspace monitoring clients.
Usage details
Who can use it: Qwen Code SDK and Managed/serve live sessions; up to 20 selected workspaces in a single batch call.
How to use it: Call the live-session snapshot batch method in the SDK and send up to 20 workspace IDs; combine the returned snapshots in a single view.
Tip: Split larger workspace sets into groups of 20 and use live updates instead of polling at short intervals.
✓ Verified by official sourceSource: Qwen Code v0.26.5 release notesKeyboard shortcuts added for Session Overview and Split View
The `Cmd/Ctrl+Shift+O` shortcut to open Session Overview and the `Cmd/Ctrl+\` shortcut to open Split View have been added to Web Shell. The shortcuts also work in Qwen Code Desktop and standalone `qwen serve` web mode.
Usage details
Who can use it: Qwen Code Web Shell, Desktop, and standalone `qwen serve`; disabled when a dialog is open or focus is in an embedded terminal.
How to use it: With a session open, press `Cmd+Shift+O` or `Cmd+\` on macOS, or `Ctrl+Shift+O` or `Ctrl+\` on Windows/Linux.
Tip: Even if your keyboard layout is not QWERTY, the physical key mapping is preserved, so press the specified physical key rather than the corresponding character.
✓ Verified by official sourceSource: Qwen Code view-navigation hotkeys PRSource: Qwen Code v0.26.5 release notesWhatsApp Business API: `conversation_analytics` field removed
Meta deprecated the `conversation_analytics` field and announced that it will no longer be available starting November 1, 2026. This change affects WhatsApp Business Account analytics requests and reporting integrations that use this field.
Usage details
Who can use it: Businesses, partners, and analytics integrations using the WhatsApp Business Platform’s Business Management API; starting November 1, 2026.
How to use it: Remove `fields=conversation_analytics` from requests; switch to the appropriate field based on your needs: `analytics`, `pricing_analytics`, `template_analytics`, `template_group_analytics`, `call_analytics`, or `group_analytics`.
Tip: Scan field names and dashboard queries in advance to avoid production errors; keep backward-compatible error handling in response parsers for the `conversation_analytics` object, which will no longer be returned.
WhatsApp Marketing Messages API: separate billing SKU for max price traffic
Starting November 1, 2026, messages sent using max price are reported with separate billing line items and identifiers from messages sent at the published rate. The Pricing Analytics API uses `MARKETING_LITE_DYNAMIC`, the Template Analytics API uses `amount_spent_max_price`, and the pricing webhook uses `marketing_lite_dynamic`; message-sending requests are unchanged.
Usage details
Who can use it: Businesses and integrations using max price in the WhatsApp Marketing Messages API; traffic covered by invoices, the Pricing Analytics API, the Template Analytics API, and the pricing webhook.
How to use it: Support the `Price type: Max Price` line item in your invoice-processing code; accept `MARKETING_LITE_DYNAMIC` in the Pricing Analytics API, `amount_spent_publish_rate` and `amount_spent_max_price` in the Template Analytics API, and `marketing_lite_dynamic` in the webhook parser.
Tip: Track published-rate and max-price costs in separate dashboard series while preserving total spend; account for a small amount of transition traffic appearing under `List` on the first November invoice.
✓ Verified by official sourceSource: Max price faturalama SKU’su: fatura ve analiz API’si güncellemeleriSource: WhatsApp değişiklik günlüğüBuilt-in code-review skill handles head drift and actionlint
The built-in code-review skill realigns after a PR head changes, runs actionlint checks, reports blockers, and stops rejected retries. This reduces the chance of automated review continuing against the wrong head.
Usage details
Who can use it: Qwen Code’s built-in code-review skill and PR review workflows.
How to use it: Run the code-review skill as usual; if the PR head changes, realignment and actionlint steps run automatically against the new head.
Tip: For CI configuration changes, check the actionlint result separately before making a review decision.
✓ Verified by official sourceSource: Qwen Code v0.26.5 release notesBuilt-in reasoning effort recognizes new Qwen prefixes
Qwen Code’s built-in reasoning effort selection now covers DeepSeek-V4 and new Qwen model prefixes. When the user provides an explicit effort value, that preference continues to override automatic detection.
Usage details
Who can use it: Qwen Code model providers; DeepSeek-V4 and new Qwen model names.
How to use it: Select the model and leave reasoning effort on automatic; if you need specific behavior, provide an explicit effort value in the provider settings.
Tip: If a new model name isn’t recognized automatically, use an explicit reasoning effort and check the values supported by the provider.
Context files already loaded are no longer imported into memory again
Qwen Code skips files already loaded as context files during memory import. This reduces unnecessary context consumption from duplicate content.
Usage details
Who can use it: Qwen Code context and auto memory loading flows.
How to use it: Load files through both context and memory in a session; the tool automatically skips duplicate files.
Tip: Keep the same file in a single source instead of manually adding it through two different mechanisms.
✓ Verified by official sourceSource: Qwen Code v0.26.5 release notes/context displays token counts of one million or more more readably
`/context` now displays token counts of one million or more in millions in terminal, plain-text, and OpenTUI views. Large context usage is quicker to read.
Usage details
Who can use it: Qwen Code CLI `/context`, plain-text, and OpenTUI views.
How to use it: Run `/context` during a session and read token counts above one million in M.
Tip: Don’t confuse the million-based display with the total limit; check the remaining context and model limit together.
✓ Verified by official sourceSource: Qwen Code v0.26.5 release notesLong lines in tool diffs wrap vertically
Diffs in Web Shell tool cards now wrap long lines in long comments or file paths instead of requiring horizontal scrolling on each line. Reviewing large diffs requires less horizontal navigation.
Usage details
Who can use it: Qwen Code Web Shell tool-card diff views.
How to use it: Open a tool diff with long lines and read the wrapped text in the card’s vertical scrolling area.
Tip: Hover over a line to see the full file path; even in the wrapped view, check all blocks before approving.
✓ Verified by official sourceSource: Qwen Code v0.26.5 release notesStartup probe distinguishes broker write failures from incorrect topic types
The Managed Agent startup probe now reports broker write failures separately from incorrect topic types. The error message provides clearer fail-fast guidance and a mqadmin remediation path.
Usage details
Who can use it: Qwen Code Managed Agent startup and broker connection diagnostics.
How to use it: Start the Managed Agent; if the probe fails, apply a fix with mqadmin based on whether the error indicates a broker write failure or an incorrect topic type.
Tip: Don’t treat the two errors as the same network issue; for a topic type error, verify the topic definition first.
✓ Verified by official sourceSource: Qwen Code v0.26.5 release notesMemory writes report skipped directories and scan limits
Memory writes now report skipped index entries separately from items skipped due to scan capacity, while preserving example filenames and warnings. The diagnostic output makes it clearer why memory is incomplete.
Usage details
Who can use it: Qwen Code auto memory write and diagnostic flows.
How to use it: After a memory write, inspect the CLI output or relevant log; evaluate omitted index and scan-cap warnings separately.
Tip: If an expected file wasn’t added to memory, check the scan-cap warning first, then narrow the scope and write again.
✓ Verified by official sourceSource: Qwen Code v0.26.5 release notesEmpty No-workspace section is now hidden in the sidebar
Web Shell doesn’t show the No-workspace sessions section in the sidebar when it loads empty. This behavior is now consistent with the empty state of the Pinned section.
Usage details
Who can use it: Qwen Code Web Shell; empty No-workspace sessions catalog.
How to use it: Open Web Shell; if the section is empty, it’s hidden automatically and reappears when a no-workspace session is created.
Tip: Use the + Add session action instead of waiting for the section to appear to add a new session.
✓ Verified by official sourceSource: Qwen Code v0.26.5 release notesGoal card and dialog localized in Russian
Russian localization has been added for Web Shell’s Goal card and Goals dialog. Untranslated strings fall back to English.
Usage details
Who can use it: Qwen Code Web Shell’s Russian interface; English fallback for missing translations.
How to use it: Set the Web Shell language to Russian and start a Goal session; the Goal card and Goals dialog will appear in Russian.
Tip: If you see a missing string, the feature remains usable; complete the command using the English fallback text.
✓ Verified by official sourceSource: Qwen Code v0.26.5 release notesThe + Add session action remains clickable when the catalog is empty
Web Shell keeps the `+ Add session` action clickable even when the session catalog is empty. The empty-state screen does not prevent you from starting a new session.
Usage details
Who can use it: An empty session catalog in Qwen Code Web Shell.
How to use it: When the catalog is empty, click `+ Add session` in the sidebar or in the empty state.
Tip: You don’t need to wait for the section to fill up to start your first session; use the action in the empty state.
✓ Verified by official sourceSource: Qwen Code v0.26.5 release notesThe No-workspace header is now aligned with other session sections
The No-workspace sessions header now uses the same height, spacing, arrow, muted text, and hover/focus behavior as the Pinned, Workspaces, and Recent sections.
Usage details
Who can use it: Qwen Code Web Shell session sidebar.
How to use it: Open the Web Shell sidebar; the No-workspace sessions section uses the same interaction pattern as the other section headers.
Tip: If the header isn’t visible, expect it to appear only once a session exists, due to the change that hides empty sections.
✓ Verified by official sourceSource: Qwen Code v0.26.5 release notesA Live WebSocket usage guide has been added to the OpenAI Go SDK
openai-go v3.91.1 adds a new guide and compilable examples explaining primary session, sideband connection, and stored-session fork flows for Live WebSocket. The guide covers typed Send/Recv usage, waiting for `session.started` after `session.start`, and consuming the `session.closed` event on shutdown; this is a documentation and examples update, not a change to API behavior.
Usage details
Who can use it: OpenAI Go SDK v3.91.1 and developers using the source repository; Live WebSocket integrations.
How to use it: Update with `go get github.com/openai/openai-go/v3@v3.91.1` and follow the repository’s `live/README.md` guide. In the primary flow, call `client.Live.Connect`, send `session.start`, and wait for the `session.started` event; the sideband and fork examples are in the same guide.
Tip: Remember that opening the WebSocket doesn’t start the session; don’t send application commands until `session.started` arrives, and consume `session.closed` events before shutdown.
✓ Verified by official sourceSource: OpenAI Go SDK v3.91.1 sürüm notlarıSource: Live WebSocket rehberi commit’i – OpenAI Go SDKOpenAI Python SDK 3.42.1 fixes WebSocket connections
Version 3.42.1 includes fixes that preserve Responses WebSocket connection behavior and validate beta Realtime WebSocket authentication profiles. It also adds SDK-managed policy helpers for Responses.
Usage details
Who can use it: All developers using the OpenAI Python SDK, especially applications using Responses WebSocket and beta Realtime WebSocket.
How to use it: Run `pip install --upgrade openai` in your virtual environment to update to 3.42.1, then rerun your WebSocket connection and authentication tests.
Tip: Before deploying to production, test Responses and Realtime WebSocket scenarios in a staging environment to verify existing connection behavior.
✓ Verified by official sourceSource: openai-python v3.42.1 sürüm notlarıOpenAI Ruby SDK 0.115.1 fixes Ruby 3.3.0 compatibility
Version 0.115.1 includes a dependency fix that prevents the use of `connection_pool` versions incompatible with Ruby 3.3.0.
Usage details
Who can use it: OpenAI Ruby SDK users, especially projects using Ruby 3.3.0.
How to use it: Upgrade the SDK to 0.115.1 with `bundle update openai` or `gem update openai`, update the lockfile, and run your Ruby 3.3.0 tests.
Tip: Pin your Ruby version and `connection_pool` dependency together in the CI matrix to catch dependency resolution issues early after the update.
✓ Verified by official sourceSource: openai-ruby v0.115.1 sürüm notlarıSession messages explain safe-mode and missing peer states more clearly
Session messaging now explains separately when messages are suppressed by safe-mode and when a peer address is missing. It no longer gives users the impression that the recipient is unavailable.
Usage details
Who can use it: Qwen Code inter-session messaging and peer connections.
How to use it: Send a session message; check the explanation shown when it doesn’t reach the recipient and the safe-mode status.
Tip: If a message doesn’t appear, check for safe-mode suppression first, then check the peer address configuration.
✓ Verified by official sourceSource: Qwen Code v0.26.5 release notesQwen Code Web Shell: Panes for sessions removed from the catalog are cleared
Web Shell removes sessions no longer in the active catalog from Split View panes. This prevents empty or stale panes linked to sessions that are no longer valid from being left behind.
Usage details
Who can use it: Qwen Code v0.26.5 and Qwen Code Desktop v0.26.5; Web Shell sessions using Split View.
How to use it: While using Split View, remove a session from the active catalog; in v0.26.5, the corresponding pane is cleared automatically.
Tip: Remove sessions you no longer use from the catalog to keep the Split View list tidy.
✓ Verified by official sourceSource: Qwen Code v0.26.5 sürüm notlarıSkill slash commands retain argument hints while typing
Argument hints for skill slash commands remain visible while typing. As the user enters arguments, the hint narrows to show only the missing arguments.
Usage details
Who can use it: Skill slash commands in Qwen Code Web Shell.
How to use it: Start typing a skill slash command and enter the arguments in order; the remaining required arguments are shown in the composer.
Tip: Fill in the missing arguments shown in the hint; follow the suggested order rather than entering a long description before completing the command.
✓ Verified by official sourceSource: Qwen Code v0.26.5 release notesNested scrollbars removed from long tool approvals
The floating card opened for tool approval in Web Shell now uses a single scroll area for long diffs. Long lines and changes are easier to scroll through.
Usage details
Who can use it: Web Shell tool approval cards, especially for operations that display long diffs.
How to use it: For a tool call with a long diff, open the approval card and use the single scroll area.
Tip: Review the entire diff in the single scroll area before approving.
✓ Verified by official sourceSource: Qwen Code v0.26.5 release notes
October 2026
Fixed file uploads from outside the workspace in `--restricted` sessions
In `--restricted` sessions, uploads from outside working directories are now blocked, even if a Read permission rule covers external paths.
Usage details
Who can use it: Claude Code sessions using `--restricted`.
How to use it: Start the session with `claude --restricted` and upload files only from within working directories; external paths will now be rejected.
Tip: Test access to a file in a parent directory outside the workspace to verify your restricted session policy.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notlarıOutput Watermarking enabled by default for Grok API text responses
Starting November 7, 2026, xAI will add an invisible statistical watermark to the text output of seven supported Grok models. The feature will be enabled by default across all API endpoints, including the Responses API and Chat Completions API; the request format is unchanged, and you can check the response’s watermark status in the x-watermark header.
Usage details
Who can use it: Starting November 7, 2026, across all xAI API endpoints, including global https://api.x.ai and regional endpoints, for the seven listed text models. Team admins can disable it per model in Console, though some models may be locked for the team.
How to use it: No code changes are required. Read the x-watermark header in a Responses API or Chat Completions response; a value of true indicates that a watermark was applied. To disable it, open Team settings > Model output watermarking in Console, deselect the relevant models using the settings icon, click Save, and then confirm the Opt Out step; changes apply to new requests within approximately 5 minutes.
Tip: Store the x-watermark header with the request ID and model in your compliance records; don’t assume the opt-out status is the same across different teams or models.
✓ Verified by official sourceModels: grok-4.7, grok-4.6, grok-4.5, grok-4.3, grok-4.20-reasoning, grok-4.20-non-reasoning, grok-4.20-multi-agentSource: Output Watermarking | SpaceXSI DocsQwen is coming to the China version of Apple Intelligence
Apple announced that it will add Alibaba’s Qwen support to select parts of the Apple Intelligence experience in China. Users will be able to use Qwen’s text and image understanding capabilities in Siri, Writing Tools, and Visual Intelligence; ask Siri questions about photos, PDFs, and presentations; and generate original content in Writing Tools.
Usage details
Who can use it: It will be tested in the iOS developer beta on supported iPhones in China starting November 4, 2026, and will roll out to users with iOS 27.3 in December 2026. Siri and the device language must be set to English or Simplified Chinese; other platforms will be supported after the necessary approvals.
How to use it: Install the relevant iOS developer beta on a supported iPhone and enable Apple Intelligence. When the Qwen experience appears, sign in with a Qwen account or use the Sign in with Apple option; then invoke it from Siri, Writing Tools, or Visual Intelligence.
Tip: Starting questions about photos, PDFs, or presentations through Siri, and rewriting or generating content from scratch in Writing Tools, makes it easier to use Qwen’s new integration surfaces separately.
✓ Verified by official sourceSource: Apple: Apple Intelligence Çin’de kullanıcılara geliyorPlan restoration after `/clear` or `/resume` fixed
Fixed a rare bug where resuming a conversation after using `/clear` or `/resume` while part of the conversation was still being saved would not restore the plan.
Usage details
Who can use it: All users who resume planned sessions in Claude Code.
How to use it: Update Claude Code; resume a conversation after `/resume` or `/clear` in a planned session, then check the plan status.
Tip: Exporting critical plans at the end of a session provides extra protection against rare save issues.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notları`claude -p` reports the reason more accurately when Auto mode stops
Fixed an issue where `claude -p` text output did not state the reason when an Auto mode turn stopped while a background agent was still running.
Usage details
Who can use it: CLI and automation workflows using `claude -p`.
How to use it: Update Claude Code and run an Auto mode workflow with `claude -p`; read the standard text output when it stops.
Tip: In CI scripts, save the output and classify an Auto mode safety stop separately from an actual process error.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notları`claude auth login` works in managed environments that require a gateway
Fixed an issue where the `claude auth login` command was rejected when managed settings required a Claude apps gateway. The terminal now displays the gateway sign-in flow.
Usage details
Who can use it: Team or Enterprise-managed Claude Code environments.
How to use it: Run `claude auth login` in the terminal and complete the displayed Claude apps gateway sign-in steps.
Tip: If your organization uses a gateway, use the organization’s gateway sign-in flow rather than a personal API key.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notlarıAutomatic action review and tool switching added to Agent sessions
The `review` setting and session tools can now be specified when creating or updating an Agents API session. `auto_review` automatically reviews eligible actions; tools changed with `agent.tools` take effect on the next model call or, for built-in tools, the next turn.
Usage details
Who can use it: The Agents API beta; available with Python SDK 3.42.0 and Node SDK 7.46.0.
How to use it: Use `review: "auto_review"` in a session request; when updating an existing session, send the `agent.tools` array to change its tools. Use `review: "none"` to turn off automatic review.
Tip: Start with `auto_review` for actions with external effects, and add only the tools needed for new turns to the session.
Background context compaction is now configurable in the Agents API
The Python and Node SDK versions dated October 31 added a setting for choosing how agent context is compacted. `compaction.async` uses background compaction by default; setting it to `false` runs compaction synchronously.
Usage details
Who can use it: Developers using the Agents API beta and OpenAI Python SDK 3.42.0 or Node SDK 7.46.0 and later.
How to use it: Update the SDK; send `compaction: { async: true }` or `compaction: { async: false }` in an agent creation or update request. Leaving the field empty or setting it to null enables background compaction by default.
Tip: For long sessions, `async: true` may reduce latency, but monitor usage metrics because background compaction may incur additional token usage.
Subagents no longer switch to the parent model by mistake when they wake up
Fixed a bug that caused a subagent to switch back to the parent model when a background task or message woke it up, even though it had inherited its model from its parent subagent.
Usage details
Who can use it: Claude Code subagents and background tasks.
How to use it: Update Claude Code; wake a subagent with an explicitly specified model using a background task, then check the model in the session transcript.
Tip: Explicitly specifying the model name in the task description for critical subagents makes inheritance behavior easier to observe.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notlarıDefault effort level for subagents can now be configured
The CLAUDE_CODE_SUBAGENT_EFFORT environment variable sets the effort level subagents use when no other setting is specified. This means subagents do not have to automatically inherit the main session’s level.
Usage details
Who can use it: Developers and team administrators using Claude Code subagents.
How to use it: Set the CLAUDE_CODE_SUBAGENT_EFFORT variable to your preferred effort level before launching Claude Code; if a level is specified separately when calling a subagent, that level takes precedence for that call.
Tip: Try a lower level for subagents to reduce cost and improve speed, and a higher level for complex tasks, on a per-project basis.
✓ Verified by official sourceSource: Claude Code değişiklik günlüğüAnthropic cloud tasks can load Claude.ai plugins
Scheduled tasks running in Anthropic cloud now load the plugins enabled in the claude.ai account. This helps tasks run more consistently with the user’s active plugin configuration.
Usage details
Who can use it: Users with Scheduled tasks running in Anthropic cloud; plugins must be enabled in the claude.ai account, and organization policy must allow them.
How to use it: Enable the plugins you want in your Claude.ai account, then create or run a Scheduled task in Anthropic cloud.
Tip: Before running the task in production, verify that the enabled plugins are available in the cloud session with a small test task.
✓ Verified by official sourceSource: Claude Code değişiklik günlüğüArtifact database access rules are displayed correctly
Fixed an issue where database access rules in the Artifact tool appeared as `?` at the `self` level. If there is a risk that rules could be deleted during publication, the rules are now quoted correctly when republishing is rejected.
Usage details
Who can use it: Claude users who use the Claude Code Artifact tool.
How to use it: Update Claude Code, then open the flow to republish an Artifact or review its database access settings.
Tip: Check the Artifact’s `self` access rules before publishing; don’t leave broad permissions enabled by default.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notlarıDefault rule behavior changed for the Auto mode allow list
When the autoMode.allow list does not include $defaults, it no longer removes the built-in rule from scheduled task prompts. This helps preserve more consistent security behavior in scheduled tasks.
Usage details
Who can use it: Claude Code users and administrators who use Auto mode and scheduled tasks.
How to use it: Check the autoMode.allow list in managed settings; explicitly use $defaults to preserve built-in rules, then test with a scheduled task.
Tip: When adding custom allow rules, verify with a test task that the default security rule has not been unintentionally disabled.
✓ Verified by official sourceSource: Claude Code değişiklik günlüğüInput tagging support added for Bedrock guardrails
For Bedrock upstreams, the Claude apps gateway option guardrail.input_tagging wraps prompts in the input tags expected by the Amazon Bedrock guardrail prompt-attacks filter. The all_messages_and_system option also tags the system prompt.
Usage details
Who can use it: Claude apps gateway administrators who use Amazon Bedrock guardrails.
How to use it: Enable guardrail.input_tagging in the relevant Bedrock upstream configuration; select all_messages_and_system to include the system prompt as well.
Tip: Run separate test requests with a regular user prompt and a system instruction to verify which message sections the guardrail scans.
✓ Verified by official sourceSource: Claude Code değişiklik günlüğüAuto-compact context warning appears for unknown model names
When a model unknown to this Claude Code version is selected with `/model <id>`, users are now notified that auto-compact will assume a 200K context window.
Usage details
Who can use it: Claude Code users with custom model names or gateway deployments.
How to use it: Run `/model <id>` during a session; note the 200K context warning that appears for an unrecognized model.
Tip: Verify the correct context window for custom model deployments; an incorrect assumption can cause compaction or limit issues.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notları`/diff` refreshes run fewer and less frequent Git commands in large repositories
In large repositories, `/diff` refreshes generally run one fewer full-repository Git command. If Git is slow, delays are added between refreshes to reduce UI load.
Usage details
Who can use it: Users running Claude Code `/diff` in large Git repositories.
How to use it: Update Claude Code, run `/diff` in a large repository, and refresh the file changes.
Tip: In very large repositories, reviewing only the relevant files or subdirectory is still faster.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notlarıCallable API keys are now scoped per request
Python SDK 3.42.0 scopes keys to each request when using callable API keys. This prevents one request’s credentials from being carried over to another in calls made through the same client.
Usage details
Who can use it: API developers using OpenAI Python SDK 3.42.0.
How to use it: Update the SDK with `pip install --upgrade openai`; pass the callable API key provider to the relevant request rather than making a persistent client-wide mutation, and test that each request uses the expected key.
Tip: In multi-tenant applications, select keys in the request context and test for key leakage when sharing a client across users.
✓ Verified by official sourceSource: OpenAI Python SDK v3.42.0 sürüm notlarıClaude Code no longer leaves shell commands running in the background when it exits
When Claude Code exits on its own on macOS and Linux—for example, when credentials are rejected or a cloud session is archived—it now terminates open shell commands.
Usage details
Who can use it: macOS and Linux users running Claude Code 2.1.316.
How to use it: Update Claude Code and use it normally; no separate setting needs to be enabled.
Tip: After authentication or cloud session errors, also review the output of `ps` and `jobs` to check for processes left running in the background.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notlarıClaude Tag can edit a shared Slack canvas
Fixed an issue that prevented Claude from adding to or editing a canvas from the Slack channel where it was created after the canvas was shared with another channel or DM.
Usage details
Who can use it: Enterprise Claude Tag organizations connected to Slack.
How to use it: Update Claude Tag, share the canvas with another channel or DM, then request an edit from the Slack channel where it was created.
Tip: Before expanding canvas sharing, test a small edit from the source channel.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notlarıParticipation setting renamed in Claude Tag
The Respond automatically toggle in a Slack channel has been replaced with the How Claude participates option on the channel’s Configure page and in Claude Tag admin settings. This makes it clearer how Claude can participate in the channel.
Usage details
Who can use it: Enterprise organizations using Claude Tag connected to Slack.
How to use it: Open the Slack channel’s Configure page or Claude Tag admin settings, then choose a participation mode from How Claude participates.
Tip: When changing automatic response behavior for a channel, also check the relevant Claude Tag spend limit and access policies.
✓ Verified by official sourceSource: Claude Code değişiklik günlüğüCloud tasks now fail correctly when Auto mode has no decision
Fixed an issue where cloud sessions and cloud scheduled tasks were not marked as failed when Auto mode stopped a turn because it received no safety decision from the server.
Usage details
Who can use it: Users of Claude Code cloud sessions and cloud scheduled tasks.
How to use it: Run a cloud session or scheduled task; if Auto mode cannot make a safety decision, the status in the run history and the notification will now show the failure.
Tip: When monitoring automated tasks, check the run status and error notification as well as the final message.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notlarıSeverity card added to Code Review analytics
The Code Review analytics page now includes a card showing total findings for the period, a daily stacked chart by severity, and the count at each level.
Usage details
Who can use it: Teams using Claude Code Code Review analytics.
How to use it: Open the Code Review analytics page and review the period and daily breakdown in the findings-by-severity card.
Tip: Prioritize based not only on total findings but also on the daily trend in high-severity findings.
✓ Verified by official sourceSource: Claude Code değişiklik günlüğüCode Review results are now shown in a findings card
In cloud and claude remote-control sessions, /code-review launched from Claude Desktop or the web Code tab now shows results in a findings card instead of a plain-text list.
Usage details
Who can use it: Claude Desktop and the web Code tab; cloud and Remote Control sessions.
How to use it: Run /code-review in the Code tab; when the review is complete, select the findings card that appears.
Tip: Verify findings one by one from the card; also open the relevant diff rather than relying only on the text summary.
✓ Verified by official sourceSource: Claude Code v2.1.316 GitHub sürüm notlarıdsh-superagent research prototype released for DeepSeek Harness
DeepSeek Harness Team released the dsh-superagent research prototype under the MIT license. It proactively handles tasks in a single GitHub repository, running permitted requests in its own session and working directory and relaying results to the source conversation, for example as a pull request. The README specifically notes that it is neither production-ready nor an official DeepSeek product.
Usage details
Who can use it: Publicly available on GitHub; can be run from a GitHub repository as a DeepSeek Harness bundle and can listen for GitHub events and optional Feishu chats. This is research/reference code and should not be considered production-ready.
How to use it: Clone the repository and ask the coding agent in the DeepSeek Harness environment: `Read SETUP.md in this repository and set up dsh-superagent with me.` During setup, select only GitHub or Feishu as the source, specify the `owner/name` repository to work on and a separate agent account, keep the allowlist narrow, and run setup after approving the plan. SETUP.md requires Node.js 22.19+ or 24+, pnpm, git 2.32+, tmux, Python 3.8+, GitHub CLI 2.48+, DSH, and a model provider.
Tip: For your first test, use SETUP.md's default `observe` acceptance mode, a separate GitHub account, and keep automatic merging disabled. Do not expand to a live repository until you have verified the permissions and agent output.
Enterprise members can manage network secrets in cloud environments
Enterprise members can now add and delete network secrets in their own cloud environments, in addition to viewing them. Previously, these actions were limited to users with administrator access.
Usage details
Who can use it: Enterprise members who use Claude Code cloud sessions; organization policies may restrict this feature.
How to use it: Open the cloud environment settings and add a new secret or delete an existing one in the network secrets section.
Tip: Keep secret values only in the network secrets section rather than pasting them into session text, and regularly remove those no longer needed.
✓ Verified by official sourceSource: Claude Code değişiklik günlüğüExisting evaluations on the Evals platform are now read-only
As the next phase of the planned retirement, existing Evals evaluations switched to read-only mode on October 31; they can no longer be edited. The Evals dashboard and API are scheduled to shut down on November 30, 2026.
Usage details
Who can use it: API developers using Evals can read existing evaluations but cannot edit them; access to the platform’s dashboard and API will continue until the planned shutdown on November 30.
How to use it: Review existing results and manually migrate the evaluation to a Promptfoo configuration; then run it with `promptfoo eval -c promptfooconfig.yaml --no-cache` and open the results with `promptfoo view`.
Tip: Before shutdown, recreate the graders, test data, and expected outputs, then compare the results; scores from manually migrated graders may not match exactly.
✓ Verified by official sourceSource: OpenAI API deprecations — Evals platformuSource: OpenAI Evals’tan Promptfoo’ya geçiş rehberiReach metric reportedly being removed from Facebook analytics
A secondary report says Meta will remove the Reach metric from Facebook analytics starting October 31, 2026, and make Viewers the primary unique audience metric. Meta’s current Help Center page highlights Views, Content interactions, and Follows in Insights, but I could not find an official announcement confirming this dated removal.
Usage details
Who can use it: Users tracking Facebook Page and creator analytics through Meta Business Suite; the affected reporting surfaces may vary by account and content type.
How to use it: Go to Meta Business Suite > relevant Facebook Page > Insights and check for Views, Viewers, or Unique Viewers fields in place of Reach. In external reporting dashboards, map legacy Reach fields to the new metrics.
Tip: Export historical Reach data and date ranges; don’t compare Viewers or Views directly with Reach as if they were the same series.
Gateway now validates all requested models against allowlists
Claude apps gateway checks every model named in a request against the `availableModels`, `deniedModels`, and `models` fields. If a model is blocked, it now returns a 400 error identifying the field causing the issue.
Usage details
Who can use it: Claude apps gateway administrators and applications sending requests through the gateway.
How to use it: Check the model fields in your gateway configuration and match all model names in the request against the allowlist.
Tip: Before deployment, run a negative test with a blocked model for each model field to verify the field name in the 400 response.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notlarıLayered merge mode added for gateway policies
The managed.policy_mode: layered setting applies all policies that match a developer and merges them, with lower-priority policies taking precedence. Unlike the previous behavior, which stopped at the first match, multiple rules now take effect.
Usage details
Who can use it: Team and Enterprise administrators managing layered policies in the Claude apps gateway.
How to use it: Set managed.policy_mode to layered in the gateway’s managed policy configuration, then check the policy order from the lowest-priority rule upward.
Tip: Review conflicting permissions with a small test group first; clearly document which fields the higher layer overrides.
✓ Verified by official sourceSource: Claude Code değişiklik günlüğüShared monthly budget introduced for groups
The Claude apps gateway now supports a single monthly budget shared by group members, in addition to per-person limits. This budget is configured using the pooled_budget field in the spend limits management API.
Usage details
Who can use it: Organization administrators who manage the Claude apps gateway and enforce group-based spending limits.
How to use it: Add the pooled_budget value to the relevant group record in the spend limits management API; members’ individual limits remain in place separately.
Tip: Track the shared budget separately from per-person limits; reassess the rate at which it is used when group size changes.
✓ Verified by official sourceSource: Claude Code değişiklik günlüğüThe free period for Search/Fetch in Harness, Claude Code, and Codex is ending
The free period for server-side Search/Fetch calls by DeepSeek API users through DeepSeek Harness, Claude Code, and Codex ends on October 31, 2026, at 16:00 UTC; per-request billing will begin after that threshold. Direct Search and Fetch API calls cost $0.75 per 1,000 successful requests.
Usage details
Who can use it: Users who use the DeepSeek API and call Search/Fetch through DeepSeek Harness, Claude Code, or Codex are affected. In Istanbul time, the threshold is October 31, 2026, at 19:00; direct API endpoints are already subject to billing rules.
How to use it: No settings need to be enabled; continue using your existing integrations. For direct use, include a DeepSeek API key in calls to `POST https://api.deepseek.com/v1/tools/search` or `POST https://api.deepseek.com/v1/tools/fetch`, and check your account balance before and after the threshold.
Tip: Cache results in agent workflows instead of repeatedly searching for or fetching the same page. Only successful requests that return results are billed; empty results, errors, and timeouts are not billed.
✓ Verified by official sourceSource: DeepSeek-V4.1-Flash-1017 resmî duyurusuSource: Search & Fetch API kullanım kılavuzuAuto mode server in the host app is enabled even behind a proxy
`CLAUDE_CODE_AUTO_MODE_SERVER=1` is no longer ignored in sessions started from a host app such as Claude Desktop behind a proxy or LLM gateway when experimental betas are disabled.
Usage details
Who can use it: Users running Claude Code through a host app, proxy, or LLM gateway.
How to use it: Set `CLAUDE_CODE_AUTO_MODE_SERVER=1` in the environment, update Claude Code, and start a new session from the host app.
Tip: If your organization’s proxy filters experimental beta headers, also review the connection logs for the Auto mode server.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notlarıTransparent markup overlays can be used in Images API edits
A transparent PNG markup overlay can be added to image editing requests to guide edits by placing it over the first image. The overlay must have the same dimensions as the first image and cannot be used with a mask.
Usage details
Who can use it: With OpenAI Python SDK 3.42.0 and Node SDK 7.46.0, in editing workflows supported by GPT Image 2.5 and 2.6.
How to use it: Update the SDK and add a transparent PNG as the `markup` file in the image edit request; prepare the PNG with the same oriented dimensions as the first image and do not send a `mask` in the same request.
Tip: Clearly mark only the area to be edited in the markup using semi-transparency; first test how the model interprets the overlay with a small sample image.
✓ Verified by official sourceModels: GPT Image 2.5, GPT Image 2.6Source: OpenAI Python SDK v3.42.0 sürüm notlarıSource: OpenAI Node SDK image markup değişikliğiSource: OpenAI API: Görüntü oluşturma ve düzenlemeK2 series closed to new API users starting October 31
Starting October 31, 2026, K2 series models are no longer available to newly registered users; existing users are unaffected. Moonshot plans to retire the entire K2 series by the end of December 2026.
Usage details
Who can use it: API accounts created after October 31 cannot use Kimi K2.7 Code, Kimi K2.7 Code Highspeed, or Kimi K2.6; users who registered earlier are unaffected until the planned retirement date.
How to use it: K2 models cannot be selected in new API accounts; switch to the Kimi K3 or Kimi K3.1 Pro model IDs for new integrations. Migrate existing K2 integrations before the end of December.
Tip: During migration, retest the model ID, context limit, and thinking/streaming parameters in your client code against the K3 or K3.1 Pro documentation.
✓ Verified by official sourceModels: Kimi K2.7 Code, Kimi K2.7 Code Highspeed, Kimi K2.6Source: Kimi API model çıkarım fiyatları — K2 erişim duyurusuSource: Kimi API model listesiExternal key management for AWS, GCP, and Azure added for organizations
Organization external key records can be created, listed, retrieved, validated, and deleted. The API also supports filtering records by purpose and listing them with pagination.
Usage details
Who can use it: Organization administrators; projects with Admin API permissions for AWS, GCP, and Azure external keys.
How to use it: Upgrade to Python SDK 3.42.0 or Node SDK 7.46.0 and start creating a record with `client.admin.organization.externalKeys.create(...)` in the Admin API client; use retrieve, list, validate, and delete calls for its lifecycle.
Tip: Validate the provider configuration with a `validate` call before activating the key, and use the `purpose` filter in lists.
MCP form requests appear as cards in cloud sessions
When an MCP server requests a form, the cloud session now displays a form card instead of silently canceling the request. Users can fill out the card or decline the request.
Usage details
Who can use it: Claude Code cloud session users whose MCP servers request forms.
How to use it: Fill in and submit the fields in the form card that appears during a cloud session; if you do not want to proceed, decline the request through the card.
Tip: If an authentication or configuration form appears unexpectedly, check the MCP server’s source and the requested fields before submitting.
✓ Verified by official sourceSource: Claude Code değişiklik günlüğüOpenAI Go SDK: Live Decision fields are preserved in JSON conversions
openai-go v3.91.0 preserves the resolved additional fields of Live Decision items in JSON conversions; tool call IDs, names, arguments, outputs, and unknown item types aren’t lost. Explicitly provided additional-field overrides, null, and Omit values retain their precedence, reducing data loss when reserializing typed Live events.
Usage details
Who can use it: Live WebSocket integrations using OpenAI Go SDK v3.91.0; Decision items in primary, sideband, and stored-session fork flows.
How to use it: Update the Go module with `go get github.com/openai/openai-go/v3@v3.91.0` and use typed DecisionItem events from the Live connection. If you want to preserve forward-compatible fields, don’t filter additional fields during JSON round-trips; use the SDK’s additional-field parameters for intentional overrides.
Tip: Preserve `RawJSON` and additional fields in test snapshots for compatibility with future item types; if you want to deliberately remove a field, filter it explicitly.
✓ Verified by official sourceSource: OpenAI Go SDK v3.91.0 sürüm notlarıSource: Live Decision öğe alanlarını koruma commit’i – OpenAI Go SDKFixed exit code 2 blocking in PowerShell hooks
Fixed an issue that prevented PowerShell hooks returning exit code 2 from blocking when a command started with `type cast`, `param`, `using`, or an attribute.
Usage details
Who can use it: Windows users running Claude Code hooks with PowerShell.
How to use it: Update Claude Code and run your existing PowerShell hook with the same exit code 2 flow; no additional configuration is required.
Tip: Test blocking behavior in security hooks using both the exit code and the error log.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notlarıRealtime WebSocket delivery reporting and typing strengthened
The Python and Node SDKs provide stronger typing for Realtime references and improve WebSocket delivery reporting and recovery. This makes it more reliable to track which session events belong to in voice applications and how state is maintained during connection drops.
Usage details
Who can use it: Realtime developers using OpenAI Python SDK 3.42.0 and Node SDK 7.46.0.
How to use it: Update the SDKs and parse WebSocket events using the SDK’s typed Realtime event types; log delivery and reconnection status, and implement resynchronization for failed events.
Tip: Persist each event’s session and event IDs in logs; after a network drop, first read the latest state from the server instead of blindly resending.
Hosted tool usage is reported in Responses output
An optional `tool_usage` field, separate from model token usage, has been added to stable and beta Responses. It carries input/output token details for image generation, the number of billable requests for web search, and cache details if available.
Usage details
Who can use it: Developers using the stable or beta Responses API with Node SDK 7.46.0; responses where tool usage is available.
How to use it: Update the Node SDK with `npm install openai@7.46.0` and read the `response.tool_usage` field on the Responses object; track `image_gen` and `web_search.num_requests` separately from the model’s `usage` field.
Tip: For cost reconciliation, compare `tool_usage` values with billing records; do not assume a missing field means zero usage.
✓ Verified by official sourceSource: OpenAI Node SDK v7.46.0 sürüm notlarıSource: OpenAI Node SDK barındırılan araç kullanımı değişikliğiMCP on/off state preserved when the runner changes machines
Fixed an issue where an MCP server’s on/off switch state was lost when a session moved to a new machine on a self-hosted runner.
Usage details
Who can use it: Organizations using MCP servers on a self-hosted runner.
How to use it: Update the runner, move a session to another machine, and verify the MCP server’s status via `/mcp`.
Tip: Also define MCP status explicitly in the configuration file in production; don’t rely only on the UI state.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notlarıSDK `register_repo_root` now loads the external directory’s CLAUDE.md file
Fixed an issue where the SDK failed to load a directory’s `CLAUDE.md` file when a `register_repo_root` request was made for a directory outside the working directory while `CLAUDE_CODE_ADDITIONAL_DIRECTORIES_CLAUDE_MD=1` was active.
Usage details
Who can use it: Developers using Claude Agent SDK who read instructions from additional directories.
How to use it: Enable the environment variable, call `register_repo_root` in the SDK, and provide the additional directory’s `CLAUDE.md` file at the start of the session.
Tip: Use a distinctive, harmless rule in an SDK test to verify that additional instructions are applied.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notlarıNo stale response sent after Stop on self-hosted runner
Fixed a race condition that could cause a message sent while Claude was running to be answered incorrectly after Stop, once the session process had changed.
Usage details
Who can use it: Organizations using Claude Code self-hosted runner.
How to use it: Update the runner and, in an active session, press Stop and verify that it has stopped before sending a new message.
Tip: In workflows, starting a new task in a new session instead of the same session after Stop reduces confusion.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notlarıDTMF key sending and acknowledgment added for SIP connections
SIP sideband connections can now send telephone key events with the `transport.dtmf.send` command, and successful sends produce a `transport.dtmf.sent` acknowledgment. Legacy notification types remain supported; the feature does not apply to WebRTC connections.
Usage details
Who can use it: OpenAI Realtime API SIP sideband connections only; WebRTC is not included.
How to use it: Send the `transport.dtmf.send` command on the SIP sideband channel, optionally associating it with an event, and handle the corresponding `transport.dtmf.sent` event.
Tip: In phone menu flows, keep a correlation ID for each key and don’t proceed to the next step until the `transport.dtmf.sent` acknowledgment arrives.
Unlabeled Claude replies no longer go missing in Slack Connect
Fixed an issue that left unlabeled replies unanswered in Slack Connect threads where Slack delivered the reply to the other organization’s app when Claude was installed in both organizations.
Usage details
Who can use it: Slack Connect threads with Claude Tag installed and Enterprise organizations.
How to use it: Update Claude Tag, then send an unlabeled follow-up message in a Slack Connect thread shared by two organizations.
Tip: When Claude is installed on both sides, use channel notes to track which organization’s app replies in the thread.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notlarıApproval and validation flow changed for server-managed settings
In Team and Enterprise sessions, server-managed settings retrieved from the Anthropic API are now applied without prompting for approval. If a stored copy cannot be validated, it is rejected rather than starting a session without applying the policy.
Usage details
Who can use it: Claude Code users who use Anthropic Team or Enterprise sign-in and server-managed settings.
How to use it: Sign in with a Team or Enterprise account and wait for managed settings to be retrieved. If there is a connection or validation issue, check the integrity of the locally stored copy and the management server.
Tip: Include the source and signature/validation status of managed settings in your organization’s deployment process; do not try to bypass a validation error by copying the local file.
✓ Verified by official sourceSource: Claude Code değişiklik günlüğüFallback to an older model on Vertex AI and Bedrock is fixed when a new model is unavailable
When the newest model is selected on Vertex AI or Amazon Bedrock before the account or region has access to it, fallback to an older model in the same family now works again instead of showing an error mid-session.
Usage details
Who can use it: Accounts and regions using Claude Code connections to Vertex AI or Amazon Bedrock.
How to use it: Use one of the `sonnet`, `opus`, or `haiku` aliases; Claude Code 2.1.316 will automatically fall back to an accessible older model.
Tip: To verify regional model rollouts, check the provider’s model catalog alongside the Claude Code session log.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notlarıSign in to the gateway from the VS Code chat panel
A sign-in flow for the Claude apps gateway that does not require a terminal has been added to the Claude Code chat panel in VS Code. Authentication can now be started directly from the editor interface.
Usage details
Who can use it: VS Code users connecting to the Claude apps gateway.
How to use it: Open the Claude Code chat panel in VS Code, select the gateway sign-in option, and complete the authentication flow in your browser.
Tip: When switching teams, use this flow instead of terminal-based sign-in to verify that the VS Code profile is connected to the correct gateway.
✓ Verified by official sourceSource: Claude Code değişiklik günlüğüApproved plans no longer switch to Auto mode after VS Code reload
Fixed an issue that caused a conversation whose plan had been approved with “Yes, and manually approve edits” to switch to Auto mode after a reload or startup.
Usage details
Who can use it: Sessions using plan approval in Claude Code for VS Code.
How to use it: Update the extension, approve the plan with manual edit approval, and check the permission mode after reloading the window.
Tip: If code changes are sensitive, quickly verify the permission mode and plan status after a reload.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notlarıEffort and Ultracode controls appear for `[1m]` model tags in VS Code
Fixed an issue where the Effort slider, Ultracode toggle, and model name disappeared when the model setting included a `[1m]` tag, such as `opus[1m]`.
Usage details
Who can use it: Claude Code for VS Code users; model settings tagged with `[1m]`.
How to use it: Update the VS Code extension, open the Claude panel, and use the Effort/Ultracode controls while a model tagged with `[1m]` is selected.
Tip: When using a model with 1M context, verify both the model tag and Effort level before starting a session.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notları/claude-api skill reads the Managed Agents workflow first
The Managed Agents examples in the /claude-api skill require reading the relevant Managed Agents workflow run before changing the agent. This reduces the risk of making changes without understanding the current run.
Usage details
Who can use it: Claude Code users who manage Managed Agents workflows with the /claude-api skill.
How to use it: Run the /claude-api skill in Claude Code and ask to change a Managed Agents agent; provide the workflow run details the skill requests, then apply the change.
Tip: Record the workflow’s current steps and failure points before making changes to avoid unnecessary reconfiguration of the same run.
✓ Verified by official sourceSource: Claude Code değişiklik günlüğüAgent sessions can be filtered by exact metadata matches
Session listing calls now support metadata filters based on exact string matches. Filters can be combined with the existing agent filter and pagination; up to 16 key-value pairs are supported.
Usage details
Who can use it: The Agents API beta and Node SDK 7.46.0; new sessions or metadata updates may take time to appear in filter results.
How to use it: List sessions with `client.beta.agents.sessions.list({ metadata: { team: "research" } })`; keep pagination and, if needed, the `agent_id` filter in the same request.
Tip: Design metadata values as small, stable labels, and account for the fact that new updates may not appear in search results immediately.
✓ Verified by official sourceSource: OpenAI Node SDK v7.46.0 sürüm notlarıSource: OpenAI Agents oturum metadata filtresi değişikliğiExperimental time-awareness setting added to the Agents API
The experimental `experimental.time_awareness` setting has been added to agent and session configurations. Enabled settings can be read in agent and session responses; partial updates preserve experimental fields that are not sent.
Usage details
Who can use it: Developers using the Agents API beta who accept experimental features; Python SDK 3.42.0 and Node SDK 7.46.0.
How to use it: Send `experimental: { time_awareness: true }` in an agent creation, update, or session configuration; do not include unused experimental fields in the request, because existing settings are preserved.
Tip: Because this is experimental, keep the setting behind a feature flag and verify outputs against the actual time for tasks involving time zones.
Push-to-talk hint no longer disappears too early in a session
Fixed an issue where the “Hold space to speak” hint used up its three displays too early in a session and showed a blank key when no push-to-talk key was bound.
Usage details
Who can use it: Claude Code terminal users who use push-to-talk.
How to use it: Update Claude Code; the spacebar hint will reappear for voice input, and you can check the keyboard settings in `/config`.
Tip: If you use push-to-talk often, save your selected shortcut in your own notes rather than relying on the hint.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notlarıBash tool no longer stops a script after the first command when SHELLOPTS is exported
Fixed a bug that caused scripts launched from the Bash tool to stop after only the first command when `SHELLOPTS` was exported.
Usage details
Who can use it: Bash users running Claude Code 2.1.316.
How to use it: Upgrade Claude Code to 2.1.316 or later and rerun the multi-command script from the Bash tool.
Tip: If the script’s behavior depends on the environment, use `set -x` to confirm that all commands are executed.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notlarıBash tool fixes aliases with line breaks and functions whose names start with a hyphen
The Bash tool’s shell no longer breaks or drops Bash aliases whose values contain line breaks or functions whose names start with a hyphen. This fix does not apply to Git Bash.
Usage details
Who can use it: Claude Code 2.1.316; macOS and Linux Bash environments.
How to use it: Upgrade to the new version and run a command using the Bash tool; your existing alias and function definitions will be used automatically.
Tip: If you use Git Bash, verify the same definitions separately; Git Bash is excluded from the scope of this fix.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notlarıCloud Files search displays paths from the project root
Fixed an issue that caused folders to appear as `../..` in the cloud session Files pane after Claude changed directories. Results now show paths relative to the project’s root directory.
Usage details
Who can use it: The Files pane in Claude Code cloud sessions.
How to use it: Have Claude change directories in a cloud session, search in the Files pane, and read the result path from the project root.
Tip: Copying the file path relative to the project root before reviewing or downloading helps avoid opening the wrong folder.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notlarıAn exclusions badge was added to the Changes pane in Cloud sessions
In Cloud sessions, the note indicating which files the Changes pane does not list is now shown as a badge next to the branch name, rather than as a separate line above the diff.
Usage details
Who can use it: Claude Code Cloud sessions.
How to use it: Open the Changes pane in a Cloud session; the note appears in the badge next to the branch name.
Tip: Check this badge when reviewing missing files or folders.
✓ Verified by official sourceSource: Claude Code v2.1.316 GitHub sürüm notlarıCatching up indicator added to Cloud and Remote Control sessions
When messages are updating as you return to a Cloud or Remote Control session, a Catching up… line is now displayed. This clearly indicates that the session is not yet up to date.
Usage details
Who can use it: Claude Code users who use Cloud sessions or Remote Control sessions.
How to use it: Return to a previously open cloud/Remote Control session; the Catching up… line appears automatically while syncing.
Tip: Do not make decisions based on an old message while this indicator is displayed; check the latest state after the update completes.
✓ Verified by official sourceSource: Claude Code değişiklik günlüğüFirst-run setup text is no longer cut off in narrow terminals
Fixed an issue that dropped words from the security notes and terminal setup description during first-run setup in narrow terminals.
Usage details
Who can use it: Users running Claude Code for the first time in a narrow terminal window.
How to use it: Update Claude Code and reopen first-run setup in a narrow terminal; the text will now appear in full.
Tip: Widening the terminal to at least 80 columns during initial setup makes the options easier to read.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notlarıOption added to preserve looped system certificates
CLAUDE_CODE_DISABLE_LOOPED_CROSS_CERT_DROP prevents mutually validating looped system cross-certificates from being dropped. It helps maintain connection compatibility when a custom CA chain requires one of these certificates.
Usage details
Who can use it: Claude Code environments using a custom CA and a looped cross-certificate chain.
How to use it: Set the CLAUDE_CODE_DISABLE_LOOPED_CROSS_CERT_DROP variable in the environment that launches Claude Code, then retry the connection in a new session.
Tip: Verify the CA chain before enabling this option broadly; use it only for affected installations rather than preserving unnecessary certificates.
✓ Verified by official sourceSource: Claude Code değişiklik günlüğüThe Enterprise survey card now waits for Enter after a 1 or 0 response
Fixed an issue where entering `1` or `0` on the Enterprise survey card was mistakenly submitted immediately after a short pause. You now need to press Enter to submit the typed response.
Usage details
Who can use it: The Claude Code survey card shown to Enterprise users.
How to use it: Type `1` or `0` on the survey card and press Enter to submit your response.
Tip: To confirm that your response was saved, wait for the card to close or for a confirmation message after pressing Enter.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notlarıGemini CLI nightly improves model selection and extension installation
Version v0.67.0-nightly.20261031 preserves the selected model after a temporary fallback, cancellation, or rewind. It also respects .git/info/exclude and core.excludesFile rules in the shadow repo, locks concurrent extension installations, and recognizes the active/focus keys for custom themes.
Usage details
Who can use it: Only in the v0.67.0-nightly.20261031 preview release; there is no indication that it has been extended to the stable Gemini CLI release yet.
How to use it: Update to the nightly release; select a model via /model and use the active and focus keys in your custom theme configuration.
Tip: Test these nightly changes in a separate environment before deploying them to production, and pin the version tag in automated installations.
✓ Verified by official sourceSource: Gemini CLI v0.67.0-nightly.20261031 sürüm notlarıGemini CLI nightly: file path, streaming, and custom theme fixes arrive
The October 31 nightly release respects .git/info/exclude and core.excludesFile rules in shadow repos, canonicalizes symbolic link paths, and handles the zero-device stat condition on Windows. It also flushes pending SSE events at the end of a stream and adds the active and focus keys to the custom theme interface schema; these are separate changes from the known model selection and extension installation fixes.
Usage details
Who can use it: Gemini CLI v0.67.0-nightly.20261031.g6814e00d6; especially relevant to Windows users, repositories with symbolic links, users of shadow repos, and those defining custom themes.
How to use it: Update your Gemini CLI nightly installation to v0.67.0-nightly.20261031.g6814e00d6. If you have a custom theme schema, use the active and focus keys; you don’t need to enable any other settings.
Tip: After updating, verify file search and ignore rules in large or symbolic-link-heavy repositories, and test the secure file-reading workflow on Windows in a small test repository.
✓ Verified by official sourceSource: Gemini CLI v0.67.0-nightly.20261031 sürüm notlarıPrompts containing invisible characters submit with one Enter press
Removed the need to press Enter a second time to submit plain prompts containing invisible characters. Commands still require separate approval for security review.
Usage details
Who can use it: All users of the Claude Code terminal interface.
How to use it: Update the client, type or paste a prompt containing invisible characters, and press Enter once.
Tip: If you suspect copied text contains invisible characters, pass the prompt through a plain-text editor first to reduce unexpected formatting.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notlarıNetwork requests no longer hang on truncated or malformed compressed responses
Fixed a bug that caused some requests to wait almost indefinitely when a compressed network response was truncated or malformed.
Usage details
Who can use it: All platforms and gateways using Claude Code 2.1.316.
How to use it: Update the client, then retry requests with your current network or proxy configuration.
Tip: If you use a proxy, monitor compressed-response errors in a separate network log; if the issue persists, try temporarily disabling compression.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notlarıPlugin MCP `headersHelper` no longer breaks the root or server URL
Fixed an issue where plugin MCP `headersHelper` scripts received a malformed `CLAUDE_PLUGIN_ROOT` and server URL when an environment variable named after a credential was set to a key/value selection such as `1`.
Usage details
Who can use it: Claude Code plugins that use an MCP server.
How to use it: Update Claude Code and restart the plugin MCP server; `headersHelper` will automatically receive the environment variables.
Tip: When debugging a plugin, log `CLAUDE_PLUGIN_ROOT` and the server URL while masking sensitive information.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notlarıAuthorized retail launch of Ray-Ban Meta announced in the Philippines
Manila Bulletin reported that Meta is preparing the first authorized retail launch of Ray-Ban Meta (Gen 3) and camera-free Ray-Ban Meta Audio models in the Philippines; Vision Express has also teased the products. The report did not announce a sale date or price, and Muse will not be available at launch. The Philippines is not on Meta’s official list of supported countries, so there is no official confirmation.
Usage details
Who can use it: Philippines; an authorized retail launch is planned, but no exact sale date or price has been announced. According to the secondary report, Muse will not be available at launch.
How to use it: For now, the Philippines is not on Meta’s list of supported countries, so don’t assume orders are available through the official Meta Store; wait for an announcement from Vision Express and Meta about the local launch.
Tip: Before buying on the grey market, verify regional warranty, firmware, Meta AI, Muse, and live translation support.
! UnverifiedModels: Ray-Ban Meta (Gen 3), Ray-Ban Meta AudioSource: Ray-Ban Meta AI glasses Filipinler’e geliyorSource: Meta AI glasses desteklenen ülkelerResponses text format and native output types updated
The Python SDK has been updated to support native output types alongside Responses format updates; the Node SDK has also adapted Responses parsing to the latest text format. This makes it easier to consume new response formats without losing type information or relying on outdated parsing assumptions.
Usage details
Who can use it: Responses API developers using OpenAI Python SDK 3.42.0 and Node SDK 7.46.0.
How to use it: Run `pip install --upgrade openai` or `npm install openai@7.46.0`; parse Responses outputs with the current SDK types and test any pinned assumptions about legacy text fields.
Tip: Run contract tests with both standard and tool-using Responses outputs before deploying; don’t rely only on the plain-text field.
Typed warmup support added for Responses WebSocket
Node SDK 7.47.0 and Python SDK 3.42.0 add typed warmup support for preparing a Responses WebSocket connection in advance. The Python release also improves preservation of WebSocket cleanup cancellation.
Usage details
Who can use it: Responses WebSocket developers using OpenAI Node SDK 7.47.0 or Python SDK 3.42.0.
How to use it: Update with `npm install openai@7.47.0` or `pip install --upgrade openai`, and use the SDK’s typed warmup support when opening a Responses WebSocket session; when closing the connection, wait for cleanup cancellation to complete.
Tip: Warming up the connection before the first user request can reduce initial response latency; still add a fallback to the normal connection path if warmup fails.
✓ Verified by official sourceSource: OpenAI Node SDK v7.47.0 sürüm notlarıSource: OpenAI Python SDK v3.42.0 sürüm notlarıSlack cloud scheduled task status list now completes
Fixed an issue that left the live status list for cloud scheduled tasks shown by Claude Tag in Slack incomplete. Runs can now update the top-level messages they create.
Usage details
Who can use it: Cloud scheduled tasks monitored in Slack through Claude Tag.
How to use it: Update Claude Tag, then start a cloud scheduled task from Slack and monitor the top-level status message while it runs.
Tip: For long-running tasks, follow the top-level status message as the primary progress indicator, not thread replies.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notlarıEnvironment variable added to disable the ultrathink keyword
The CLAUDE_CODE_DISABLE_ULTRATHINK environment variable disables Claude Code’s activation of the ultrathink keyword. This lets teams disable the trigger for policy or cost reasons.
Usage details
Who can use it: Claude Code 2.1.316 users and administrators.
How to use it: Add the CLAUDE_CODE_DISABLE_ULTRATHINK variable to your environment before launching Claude Code, then start a new session.
Tip: To apply the variable only to specific projects, define it in the project’s launch script instead of the global profile.
✓ Verified by official sourceSource: Claude Code değişiklik günlüğü`/btw` history is no longer cleared prematurely in VS Code when there’s a settings error
Fixed a bug that caused saved `/btw` follow-up question history to be cleared too early when the settings file contained an error.
Usage details
Who can use it: Users of `/btw` in Claude Code for VS Code.
How to use it: Update the extension; when a settings error occurs, reopen `/btw` history in the VS Code Claude panel.
Tip: Resolve settings file errors using the warning in the panel, and save important answers rather than relying on history.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notlarıRuns with ultrathink xhigh effort in VS Code
The ultrathink keyword in VS Code now runs the relevant turn at the xhigh effort level and uses a rainbow indicator in the chat box, like in the terminal.
Usage details
Who can use it: Users of the ultrathink feature in the Claude Code VS Code extension.
How to use it: Use the ultrathink keyword in a prompt in Claude Code chat in VS Code; the turn will run with xhigh effort, and the indicator will appear in the chat box.
Tip: Use it for long, challenging turns that require reasoning; use normal effort for routine requests to avoid unnecessary consumption.
✓ Verified by official sourceSource: Claude Code değişiklik günlüğüAliases and functions transfer more reliably in Windows Git Bash
Fixed an issue where the Bash tool lost aliases and functions when the Git Bash rc file set `errexit` and `pipefail` and had only winpty aliases or no aliases.
Usage details
Who can use it: Claude Code users running Git Bash on Windows.
How to use it: Update Claude Code and rerun aliases or functions from your Git Bash rc file with the Bash tool.
Tip: To troubleshoot environment differences, compare the output of `type` in both the interactive terminal and the Claude Code Bash tool.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notlarıGrep and Glob searches that time out on Windows no longer hang
Fixed an issue where Grep and Glob would stop responding if the search process failed to exit after being killed on timeout.
Usage details
Who can use it: Users of Claude Code Grep and Glob tools on Windows.
How to use it: Upgrade Claude Code to 2.1.316 or later and run the same Grep/Glob search.
Tip: Narrowing the search scope in large repositories reduces timeouts and resource usage on Windows.
✓ Verified by official sourceSource: Claude Code 2.1.316 sürüm notlarıPermission setting blocks writes outside the working directory
The new permissions.blockWritesOutsideWorkingDirectories setting makes the Write, Edit, and NotebookEdit tools reject paths outside working directories in every permission mode. This provides an additional security layer that narrows file access for sessions running automatically.
Usage details
Who can use it: Claude Code 2.1.314 and later.
How to use it: Enable permissions.blockWritesOutsideWorkingDirectories in managed or user settings.
Tip: Explicitly define working directories and enforce this setting in CI and on shared machines.
✓ Verified by official sourceSource: Claude Code 2.1.314 sürüm notlarıChatGPT introduces stricter safety boundaries for violent content
OpenAI said it will provide fewer potentially misusable details about past mass attacks, such as weapons, equipment, acquisition, and the scale of harm. Responses may be more conservative when warning signs of violence appear in recent conversations; conversations involving an obsession with violence may be ended automatically, and accounts may be temporarily paused in repeated cases.
Usage details
Who can use it: All ChatGPT users; protections will roll out gradually over the few weeks following the announcement.
How to use it: There is no user-controlled setting; ChatGPT will apply these boundaries automatically as the rollout begins. For requests related to violence, ask about safe topics such as historical context, impacts on victims, warning signs, or prevention.
Tip: Ask for a prevention and safe-response framing instead of weapon acquisition, setup, or attack details; this makes it more likely you’ll receive useful context.
✓ Verified by official sourceSource: OpenAI: Topluluk güvenlik raporu ve gençler için yeni korumalarComputer Use tool reaches general availability
The Computer Use tool is now generally available (GA) on the Interactions API and GenerateContent API. Developers can now use Gemini-based systems in production to take actions in browser, mobile, and desktop interfaces.
Usage details
Who can use it: Gemini API users; projects using the Interactions API and GenerateContent API.
How to use it: Add the Computer Use tool from the Gemini API documentation to your request’s tool configuration.
Tip: Before deploying to production, restrict action permissions and steps requiring human approval with separate security policies.
✓ Verified by official sourceModels: Gemini 3.8 Flash, Computer UseSource: Release notes | Gemini API - Google AI for DevelopersComputer use reaches GA on Enterprise Agent Platform with HITL safety governance
Computer use is now generally available on Gemini Enterprise Agent Platform and includes comprehensive Human-in-the-Loop safety governance. Developers can configure and adjust specific approval policies when needed, while administrators can enforce organization-wide guardrails with configurableComputerUseToolSafetyPolicies.
Usage details
Who can use it: GA for Gemini Enterprise Agent Platform users; developer and administrator safety policies are available, especially for risky actions such as financial transactions and data modification.
How to use it: Configure approval policies in your application that uses the Computer use tool; apply the configurableComputerUseToolSafetyPolicies constraint for organization-wide mandatory rules.
Tip: Require approval for financial or destructive operations; use automatic approval only for low-risk, reversible actions.
DataGuard for Agent available free in public preview
Alibaba Cloud is offering DataGuard for Agent, its AI-powered database security solution, free in public preview. It can inspect supported agent inputs, tool calls, outputs, and sessions; in Guard mode, it can block supported high-risk operations.
Usage details
Who can use it: Alibaba Cloud AI workflows; public preview is free, but supported products, regions, and environments vary by feature.
How to use it: Open Alibaba Cloud’s AI-powered database security solution, select DataGuard for Agent, and configure monitoring in Observe mode first, followed by intervention in Guard mode if needed.
Tip: First measure normal agent behavior in Observe mode and filter out false positives; enable the blocking policy only afterward.
Inference metadata logging reaches GA on Gemini Enterprise Agent Platform
For each supported Gemini inference request, the platform can write a structured Cloud Logging record containing the model, region, token counts, Provisioned Throughput usage, latency, and error code. Records contain metadata only; prompt and response text are not logged.
Usage details
Who can use it: Generally available; can be used at the project or model/region level with Gemini models that use generateContent and streamGenerateContent, with Provisioned Throughput and supported pay-as-you-go options.
How to use it: In Google Cloud Console, go to Provisioned Throughput > Metrics settings, turn on Enable detailed metrics, and save. Alternatively, in the REST API, set inferenceEventLoggingConfig.state to ENABLED in the setPublisherModelConfig call.
Tip: Send custom header labels prefixed with x-goog-agent-platform-observability-label- to analyze costs and latency by team, application, or tenant.
Gemini edits an entire presentation in Google Slides with a single prompt
Gemini in Google Slides can now edit multiple slides or an entire presentation with a single prompt. It can update fonts, backgrounds, and colors in bulk; refresh content based on Drive files; combine and expand slides; and apply changes to native, editable slides.
Usage details
Who can use it: The feature is available on desktop and initially only in English. It is rolling out to Business Standard and Plus, Enterprise Standard and Plus, Google AI Pro and Ultra, and Google AI Pro for Education accounts; the expanded rollout across Rapid Release and Scheduled Release domains began on October 29, 2026. Workspace customers are getting a promotional higher limit for creating multiple slides through at least December 1, 2026.
How to use it: On desktop, open a presentation in Google Slides, click `Ask Gemini` in the upper-right corner, and describe the change you want in the side panel—for example, make all headings the same font size or summarize slides 5–7 on a single slide. Click `Submit`; when the task is complete, a summary of the changes will appear in the side panel, and you can review the edits using `Undo` or version history.
Tip: Clearly specify the scope in your prompt: include slide numbers, the desired tone or colors, and, if needed, a reference file in Drive. Make a copy of the presentation before applying major changes, and check the summary of the results.
Protections around weapons information and age prediction strengthened for teens
ChatGPT further restricts practical, potentially misusable information about weapons and security systems for users placed in the under-18 experience. OpenAI has also strengthened its age prediction system to continually assess new usage signals; adults mistakenly placed in the teen experience can verify their age through Persona.
Usage details
Who can use it: Users under 18 covered by ChatGPT for Teens and consumer accounts predicted to belong to someone under 18; the new protections will roll out gradually over the next few weeks.
How to use it: The protection is applied automatically. If an adult account is misclassified, start age verification through Persona in ChatGPT under Settings > Account.
Tip: For teen users, link a parent account and review the existing parental controls and safety notifications together.
New registration and retirement schedule announced for the K2 series
The official pricing document states that, starting October 31, 2026, K2 series models will no longer be available to newly registered users, and that all K2 models are planned to be retired by the end of December 2026.
Usage details
Who can use it: Existing users are unaffected during the transition period; K2 access will be closed to newly registered users after October 31.
How to use it: Migrate API applications using K2 to the Kimi K3 or Kimi K3.1 Pro model IDs.
Tip: Test the migration before the end of December by changing the model ID in the configuration file.
The /permissions tab has been renamed Recently blocked, and unreadable calls can’t be approved
The former review tab in /permissions is now called Recently blocked. Calls that are entirely unreadable—for example, because they’re truncated when the terminal is too short—can no longer be approved from this screen.
Usage details
Who can use it: Claude Code’s /permissions interface.
How to use it: Open /permissions and go to the Recently blocked tab; make the entire call visible before approving it.
Tip: Enlarging the terminal or opening the call’s details helps prevent accidentally approving incomplete content.
✓ Verified by official sourceSource: Claude Code v2.1.314 GitHub sürüm notları/ultrareview checks workspace trust before running Git
/ultrareview now checks whether the workspace is trusted before starting Git operations.
Usage details
Who can use it: Claude Code /ultrareview command.
How to use it: Run /ultrareview in a trusted repository; in an untrusted folder, approve the workspace first.
Tip: Before approving trust in an unfamiliar repository, inspect its Git hook and contents.
✓ Verified by official sourceSource: Claude Code v2.1.314 GitHub sürüm notları3D Make is now available in the Kimi Work desktop app
Moonshot’s official announcement dated October 30 said that 3D Make is available in Kimi Work as well as Kimi.ai. This makes it possible to use workflows for creating models, scenes, and games by connecting to Blender and CAD tools within Kimi Work; this is an additional desktop availability detail alongside the known web/Kimi app access.
Usage details
Who can use it: Kimi Work desktop users; the official product page lists Windows and macOS (Apple silicon) versions. The announcement does not specify any country or membership plan restrictions.
How to use it: Open Kimi Work and enter Work mode, then describe your 3D goal in natural language. For local Blender, CAD, or game engine workflows, install the relevant 3D plugin from the Plugins section in Work and submit your request.
Tip: Clearly specify the reference image, dimensions, and expected output type in your first prompt; then add detail through multi-turn editing.
✓ Verified by official sourceModels: Kimi K3.1 ProSource: Kimi.ai on X: Introducing 3D Make in KimiSource: How to Create Better 3D Projects with KimiSource: Kimi 3D Make: Feature GuideFrames being removed from Advantage+ catalog ads and replaced with Image templates
Meta announced that the Frames option, which adds borders to product images, will no longer be available after October 30, 2026, and will be replaced by Image templates. The new templates can combine product images, catalog fields, logos or images, text, stickers, and background layers.
Usage details
Who can use it: Advertisers using Advantage+ catalog ads; Image templates are created for each catalog in Commerce Manager and applied to single image or carousel ads in Ads Manager.
How to use it: Create a template via Commerce Manager > Catalog > Image templates > + Create. In Ads Manager, select it via Sales campaign > Product data > Ad > Products from your catalog > Ad creative > Image template > Choose template.
Tip: Prepare templates in 1:1, 4:5, and 9:16 ratios; turn off Background generation and Dynamic overlays if they aren’t needed to prevent conflicts.
A subagent’s autoCompactWindow value can now reduce the global window
A subagent’s autoCompactWindow setting can now reduce the global window set by CLAUDE_CODE_AUTO_COMPACT_WINDOW, but it can’t increase it.
Usage details
Who can use it: Claude Code subagents and automatic compaction settings.
How to use it: Set the global upper limit with CLAUDE_CODE_AUTO_COMPACT_WINDOW; keep the subagent’s autoCompactWindow value below that limit.
Tip: Use a wider window for demanding tasks and a narrower one for helper tasks to balance cost and context continuity.
✓ Verified by official sourceSource: Claude Code v2.1.314 GitHub sürüm notlarıAuto mode gives clearer notices for blocked calls
Auto mode messages now state which command was blocked, that Claude can try again, and that a security check will be performed again; they also direct you to review /permissions.
Usage details
Who can use it: Claude Code Auto mode.
How to use it: Wait for a call to be blocked in Auto mode; read the notice and follow the /permissions path if needed.
Tip: Before trying again, review whether the command is really necessary and check the permission rule.
✓ Verified by official sourceSource: Claude Code v2.1.314 GitHub sürüm notlarıIdle compaction expands to subscribers with telemetry disabled and additional providers
Idle compaction now also works for subscribers with telemetry disabled. When one-hour prompt caching is enabled, it also covers Bedrock, Vertex, Foundry, and LLM gateway sessions.
Usage details
Who can use it: Claude Code subscribers; one-hour prompt caching must be enabled for Bedrock, Vertex, Foundry, and LLM gateway users.
How to use it: Leave a conversation of over 200K idle; enable one-hour prompt caching with supported providers.
Tip: In long sessions, monitor the cache duration and the context summary after compaction, and reconfirm critical instructions.
✓ Verified by official sourceSource: Claude Code v2.1.314 GitHub sürüm notlarıAutomatic idle compaction comes to Claude Code
Conversations exceeding 200,000 tokens and left idle in VS Code and the Claude Desktop Code tab are automatically summarized. The summary is used in the message after the prompt cache expires, making it easier to continue long sessions.
Usage details
Who can use it: For users of Claude Code in VS Code and the Claude Desktop Code tab; coverage has expanded subject to provider and telemetry conditions.
How to use it: Update Claude Code; leave the Idle compaction option in /config enabled in settings.
Tip: Before pausing a long task, check context usage with /context and /usage.
✓ Verified by official sourceSource: Claude Code changelogClaude Opus 4.8 fast mode is being deprecated
Fast mode for Claude Opus 4.8 was deprecated as of October 30 and will be removed on November 30, 2026. After removal, requests with speed="fast" for this model will fail; they will not automatically fall back to standard speed.
Usage details
Who can use it: Affects developers using Opus 4.8 fast mode on the Claude API and related managed platforms; Opus 5.5 fast mode remains available.
How to use it: In /v1/messages requests, stop using model Claude Opus 4.8 with speed="fast" and switch to Claude Opus 5.5; you can use the same speed="fast" parameter with Opus 5.5.
Tip: Test the model migration and cost/response speed differences in a staging environment before November 30.
✓ Verified by official sourceModels: Claude Opus 4.8, Claude Opus 5.5Source: Claude Platform release notesSource: Fast mode belgeleriThe claude_code.safety_event OpenTelemetry event is no longer sent
Claude Code has stopped sending the claude_code.safety_event OpenTelemetry event. Consumers relying on this event need to switch to new observability signals.
Usage details
Who can use it: Claude Code integrations using OpenTelemetry export.
How to use it: Find claude_code.safety_event dependencies in your telemetry pipeline and stop treating the event that is no longer sent as required data.
Tip: Do not flag the absence of this event as an error in schema or dashboard checks.
✓ Verified by official sourceSource: Claude Code v2.1.314 GitHub sürüm notlarıA Bash editor with syntax highlighting is now available for the Cloud environment setup script
The setup script field in Cloud environment settings now offers a Bash editor with syntax highlighting, line numbers, and a Copy button.
Usage details
Who can use it: Claude Code Cloud environment settings.
How to use it: Open the setup script field in Cloud environment settings; edit the script or copy it using the Copy button.
Tip: Use the line numbers to quickly find the step that caused an error before saving the setup script.
✓ Verified by official sourceSource: Claude Code v2.1.312 GitHub sürüm notlarıCloud sessions can start despite a failed setup script
Cloud sessions started from web, desktop, and mobile apps are no longer considered complete failures when the setup script exits with an error code. The session can start, allowing the user to address the setup error afterward.
Usage details
Who can use it: Users of Claude Code cloud sessions in web, desktop, and mobile apps; organization policy must not disable the feature.
How to use it: Run your setup script as usual when starting a Cloud session; even if the script ends with an error, enter the session and check the working environment and error output.
Tip: Don’t treat the session starting as a sign of success; first verify dependencies, the working directory, and setup logs.
✓ Verified by official sourceSource: Claude Code v2.1.314 sürüm notlarıYou can now select an effort level when choosing a model for Cloud routines
An effort level has been added to the model picker when creating or editing a routine running in Cloud.
Usage details
Who can use it: Claude Code Cloud routines.
How to use it: Create or edit a routine on the Cloud Routines page; select the effort level in the model picker.
Tip: Choose low effort for regular, simple tasks and higher effort for uncertain or complex tasks.
✓ Verified by official sourceSource: Claude Code v2.1.314 GitHub sürüm notlarıLong headless and SDK runs using many subagents use less memory
In long headless and SDK runs, the main session no longer holds subagents’ tool results until the end of the run. This reduces memory consumption in processes that run many subagents.
Usage details
Who can use it: Headless Claude Code and SDK sessions; especially jobs that run many subagents.
How to use it: In headless mode, start a run with subagents using claude -p or the SDK; the change applies automatically.
Tip: For large parallel jobs, still keep the number of subagents to what each task requires.
✓ Verified by official sourceSource: Claude Code v2.1.314 GitHub sürüm notlarıA hook that permits MCP calls in Cowork cloud Auto mode no longer counts as a response
In Cowork cloud sessions, a hook permitting an MCP tool call is no longer treated as the final safety response by Auto mode. The classifier or permission prompt makes the decision.
Usage details
Who can use it: Auto mode and MCP tools in Cowork cloud sessions.
How to use it: Start an MCP call in a Cowork cloud session with Auto mode; when needed, make the decision through the classifier or permission prompt.
Tip: Use hooks only as a pre-filter; don’t assume the hook alone makes the safety decision.
✓ Verified by official sourceSource: Claude Code v2.1.314 GitHub sürüm notlarıCUA Driver v0.27.0 ships ready-to-use binaries, SDK, and MCP packages
Qwen CUA Driver v0.27.0 ships ready-to-use binaries for macOS, Linux, and Windows; the same release also offers `@qwen-code/cua-sdk` for Node.js, a browser SDK, and `@qwen-code/node-repl-mcp`. Text-based MCP payload filtering options have been added for relative coordinates and model API routes.
Usage details
Who can use it: On macOS, it includes a signed and notarized universal binary and `QwenCuaDriver.app`; on Linux, it requires an x86_64/arm64 glibc 2.31 base; on Windows, it includes an x86_64/arm64 UIAccess worker. Windows worker deployment requires signing and establishing trust on the target machine; Node.js packages are published as release assets.
How to use it: Install the v0.27.0 asset for your operating system, or add `@qwen-code/cua-sdk` to your Node.js project and, if needed, `@qwen-code/node-repl-mcp` for MCP. For relative coordinates, use `CUA_DRIVER_RS_COORDINATE_SPACE=1` and optionally `CUA_DRIVER_RS_COORDINATE_SCALE=1000`; for payload filtering on model API routes, use `MCP_MODEL_PAYLOAD_FILTER=1`.
Tip: Enable relative coordinates only if the target app’s scaling behavior requires them; on Windows, complete signing and the trust chain before deploying the UIAccess worker to production.
✓ Verified by official sourceSource: Qwen CUA Driver v0.27.0 sürüm notlarıGateway upstreams gain a routing setting for IdP groups
An optional groups list has been added to Claude apps gateway upstream definitions. This lets only developers in the specified identity provider groups be routed to the relevant upstream; the rule is also preserved during failover.
Usage details
Who can use it: Teams managing Claude apps gateway and organizations using IdP group mapping.
How to use it: Add the groups list that determines developer access to the gateway upstream configuration and match the IdP group names to your existing management configuration.
Tip: Test with authorized and unauthorized accounts to verify that the same group restriction applies during failover.
✓ Verified by official sourceSource: Claude Code v2.1.314 sürüm notları`extensions validate` command added to Gemini CLI nightly
Version v0.67.0-nightly.20261030 adds the `gemini extensions validate` command, which checks local extension configurations before installation or publication. The command checks path safety and file existence for gemini-extension.json, SemVer formatting, and MCP server transport/URL configurations; the 429 rate-limit wait time is also shown in the loading indicator.
Usage details
Who can use it: Only in the v0.67.0-nightly.20261030.gb633c0703 preview release; for Gemini CLI extension developers using the nightly channel.
How to use it: Install the nightly release and run `gemini extensions validate <extension-path>`; fix any errors before installing or publishing the extension.
Tip: Add validation as a CI step before publishing an extension to catch path traversal and invalid MCP definitions early.
✓ Verified by official sourceSource: Gemini CLI v0.67.0-nightly.20261030 sürüm notlarıSource: Gemini CLI `extensions validate` pull request’iS&P Global Adaptive Retrieval connector added to Gemini Enterprise
The S&P Global - Adaptive Retrieval data store is now available in Gemini Enterprise as a Public Preview. Users can ask natural-language questions of S&P Global data agents they are authorized to access and get answers from company financials, filings, and earnings call transcripts.
Usage details
Who can use it: Public Preview; available only in the global, us, and eu locations. Access authorization for the relevant data agents in an S&P Global account is required.
How to use it: Add an S&P Global - Adaptive Retrieval data store to an application in Gemini Enterprise and complete the official setup flow; then ask natural-language questions about data you are authorized to access.
Tip: Narrow federated search results by clearly specifying the company name, period, and expected document type in your questions.
Gemini Notebook Enterprise gets a redesign, audio language selection, and summary customization
The new interface design for Gemini Notebook Enterprise is now GA. New features include choosing an output language for audio overviews independently of the notebook language, previewing infographic styles, and changing the automatic notebook summary; Web and Fast Research have replaced the Discover sources option.
Usage details
Who can use it: Generally available in the Gemini Enterprise Standard, Plus, Pay-as-you-go, and Frontline editions.
How to use it: In a notebook, select the audio language via Settings > Output language; change the summary via More options > Customize notebook. Review style examples when creating an infographic, and use the Web or Fast Research option for web research.
Tip: Choose an audio overview language independently of the source language to create more accessible work notes for different teams.
✓ Verified by official sourceSource: Gemini Enterprise sürüm notları: Gemini Notebook Enterprise görsel yenilemesiGitHub Enterprise connection test now also verifies the path used by Cloud sessions
The GitHub Enterprise Test connection action in admin settings now also tests the second connection path used by Cloud sessions and warns if the second test fails.
Usage details
Who can use it: Admin settings for organizations using Claude Code Cloud sessions.
How to use it: Open the GitHub Enterprise section in admin settings and click the Test connection button.
Tip: Even if the first test succeeds, resolve the second connection warning before starting your Cloud session.
✓ Verified by official sourceSource: Claude Code v2.1.314 GitHub sürüm notlarıGPT-6 Rosalind Research launches in Work/Codex and the API
GPT-6 Rosalind Research brings GPT-6 reasoning to multistep life sciences research, with safeguards designed for biological work, and is available in Work, Codex, and the API to a limited number of approved Research customers. It is not available in the Chat tab; existing customers can continue using GPT-5.5 Rosalind Research, and its API identifier is `gpt-6-rosalind`.
Usage details
Who can use it: Available to a limited number of Business and Enterprise customers whose organizations and users are approved for Rosalind Research; in Enterprise, a two-week Admin Preview starts off disabled. API access is limited to internal research use by separately approved organizations; the Rosalind Academic Plan does not include API access or API credits.
How to use it: In Work or Codex, select GPT-6 Rosalind Research from the model picker; in Codex, install Plugins > Life Sciences Research Plugin if needed. For the API, after your organization and project have been separately approved for access, set the model parameter in your request to `gpt-6-rosalind`.
Tip: Instead of loading large omics or biological datasets directly into the context, point to approved storage or a database and ask targeted research questions; let the model use scientific tools and sources.
✓ Verified by official sourceModels: GPT-6 Rosalind Research, gpt-6-rosalindSource: Model Release Notes — GPT-6 Rosalind ResearchSource: ChatGPT Business yayın notları — GPT-6 Rosalind ResearchSource: ChatGPT Enterprise ve Edu yayın notları — GPT-6 Rosalind ResearchAccount switching is expanding to the iOS and Android apps
Rolling out gradually to ChatGPT mobile apps, account switching lets you keep up to two accounts signed in at once and switch between them without logging out. Each account keeps its own chats, memory, settings, and subscription; accounts are not merged.
Usage details
Who can use it: Rolling out gradually in the ChatGPT iOS and Android apps; organization policies may apply in managed workspaces.
How to use it: In the app, go to Settings > Switch account and add a second account; then select the active account from the account menu.
Tip: To keep work and personal accounts separate, check the active account before switching; chats, files, and memory are not transferred to the other account.
✓ Verified by official sourceSource: ChatGPT yayın notları — Account switching on iOS and AndroidHook if conditions match Monitor commands like Bash patterns
Hook if conditions such as Monitor(tail *) now match shell commands from the Monitor tool, similar to how Bash(tail *) patterns match Bash calls.
Usage details
Who can use it: Claude Code hook configurations and the Monitor tool.
How to use it: Define an if condition such as Monitor(tail *) in your hook settings; Monitor shell calls will match this pattern.
Tip: When adapting existing Bash patterns for Monitor, specify the tool name explicitly.
✓ Verified by official sourceSource: Claude Code v2.1.314 GitHub sürüm notlarıFlagged requests are retried with another model when eligible
If ANTHROPIC_DEFAULT_OPUS_MODEL, availableModels, or deniedModels settings leave only an older Opus model or no models at all, a flagged request is now retried with another model instead of ending in an immediate refusal.
Usage details
Who can use it: Claude Code setups, especially those using organization model allow/deny lists or a default Opus model.
How to use it: If you want to preserve the current behavior, set CLAUDE_CODE_DISABLE_REFUSAL_FALLBACK=1; otherwise, retries are enabled by default.
Tip: When configuring model restrictions, leave at least one compatible fallback model on the allow list.
✓ Verified by official sourceSource: Claude Code v2.1.314 GitHub sürüm notlarıModel switching after a flagged message now covers custom model names too
Claude Code now applies automatic model switching after a flagged message, even for unrecognized custom deployment names. To disable this behavior, use CLAUDE_CODE_DISABLE_REFUSAL_FALLBACK=1.
Usage details
Who can use it: Bedrock, Vertex, Foundry, or Mantle sessions using a custom deployment name.
How to use it: Set CLAUDE_CODE_DISABLE_REFUSAL_FALLBACK=1 to disable automatic fallback; switching is enabled by default.
Tip: Clearly document in your provider settings which model family your custom model names belong to.
✓ Verified by official sourceSource: Claude Code v2.1.314 GitHub sürüm notlarıManaged session behavior improved when MDM and remote settings conflict
Differences in base URL or apiKeyHelper between MDM and remote settings are handled more consistently when signing in to claude.ai, Team, and Enterprise.
Usage details
Who can use it: Team and Enterprise organizations using MDM or remote settings.
How to use it: Sign in with your organization account and start Claude Code; define base URL and apiKeyHelper values in your organization’s managed settings.
Tip: Explicitly defining the same value in both MDM and remote settings makes troubleshooting easier.
✓ Verified by official sourceSource: Claude Code v2.1.314 GitHub sürüm notlarıMeta Business Agent UI skills can send video messages
Meta Business Agent added the component_type=video value for UI skills. The agent can now send customers a video message with an optional caption, using an HTTPS video URL or an uploaded media ID; existing video UI skills are also listed.
Usage details
Who can use it: WhatsApp Business phone numbers using the Meta Business Agent API; integrations that can provide a video URL or uploaded video ID.
How to use it: Use an Authorization Bearer token and X-API-Version: 2.0.0 in a POST https://api.facebook.com/{entity_id}/agent-ui-skills request. In the JSON body, send component_type=video, status=enabled, title, and an instruction specifying the video’s HTTPS URL or media ID. Captions must be no longer than 1,024 characters.
Tip: Prefer an accessible HTTPS MP4/3GPP URL; check the expiration period for uploaded media IDs and the 16 MB video limit.
Execution boundary and settings protection added for MiMo Code project configuration
The change merged into the main branch on October 30 blocks MCP, LSP, and formatter execution settings from project configuration in the hosted Node engine when they fall outside the host resource boundary; Claude compatibility and symbolic links are also handled by this boundary. The same change preserves existing MCP/LSP/formatter sections, immutable env/file references, relative plugin paths, and disabled settings in `PATCH /config` updates.
Usage details
Who can use it: Available in the MiMo Code main branch and subsequent builds that include this merge; applied automatically in the hosted Node engine. Normal Bun CLI behavior is preserved; if `MIMOCODE_ONLY_CONFIG_DIR_PLUGINS` is enabled, only resources from the permitted configuration directory are run in Bun as well.
How to use it: Upgrade MiMo Code to a current version that includes this merge. No additional settings are needed for hosted Node; to enforce the boundary in Bun CLI, launch it with `MIMOCODE_ONLY_CONFIG_DIR_PLUGINS=1` and keep project plugin/MCP/LSP/formatter definitions in files under `MIMOCODE_CONFIG_DIR`. When you update a model or configuration entry with `PATCH /config`, sections filtered by the permission policy are preserved.
Tip: Keep MCP commands and other executable project settings only under a trusted `MIMOCODE_CONFIG_DIR`; rejecting paths outside it and symbolic links reduces the risk of unwanted commands being run from a project.
✓ Verified by official sourceSource: MiMo Code #2737 birleşme kaydı ve açıklamasıSource: MiMo Code 30 Ekim commit diff’iOpenTelemetry now reports the actual name of official plugin MCP servers
The OpenTelemetry mcp_server.name field now reports fixed-URL remote MCP servers in official Anthropic marketplace plugins by their actual server name instead of custom.
Usage details
Who can use it: Installations using Claude Code OpenTelemetry and official marketplace MCP plugins.
How to use it: Read the mcp_server.name field in OpenTelemetry output; no additional settings are required.
Tip: Map old custom labels in observability dashboards to the new actual names.
✓ Verified by official sourceSource: Claude Code v2.1.314 GitHub sürüm notlarıPer-session concurrency limits added for prompt and agent hooks
Prompt and agent hooks, including subagents, now run at a maximum of 24 concurrently per session. Up to 12 additional jobs are held for events waiting beyond the timeout; timeout values under one second count as one second.
Usage details
Who can use it: Claude Code hook API; subagents are also included.
How to use it: Check timeout values in hook definitions; plan for values under 1 second to be treated as at least 1 second.
Tip: For jobs that use many hooks, design short, idempotent hooks to reduce queueing.
✓ Verified by official sourceSource: Claude Code v2.1.314 GitHub sürüm notlarıShort system prompt behavior clarified for unrecognized custom model names
Claude Code uses a short system prompt for unrecognized model names on Bedrock, Vertex, Foundry, and Mantle. To retain the full prompt, set CLAUDE_CODE_SIMPLE_SYSTEM_PROMPT=0.
Usage details
Who can use it: Custom deployment names and inference profile ARNs from third-party providers such as Bedrock, Vertex, Foundry, and Mantle.
How to use it: Set the CLAUDE_CODE_SIMPLE_SYSTEM_PROMPT=0 environment variable to use the full system prompt.
Tip: If your custom deployment’s model family isn’t recognized correctly, test the short and full prompt behavior separately.
✓ Verified by official sourceSource: Claude Code v2.1.314 GitHub sürüm notlarıSingle-file read limit reduced to 100,000 characters
In Kimi Code CLI 2.4.1, the amount of a file the tool can read at once is capped at 100,000 characters, or approximately 25,000 tokens. Users who want a higher limit can increase it again through configuration.
Usage details
Who can use it: Kimi Code CLI 2.4.1 users.
How to use it: Increase the max_chars value in the [read] section of config.toml.
Tip: For large log or data files, increase max_chars carefully to preserve context instead of splitting the file into chunks.
✓ Verified by official sourceSource: Kimi Code CLI ChangelogSelf-disclosure label for AI content added to the Threads API
Meta added support for disclosing AI-generated content in text, image, video, and carousel posts created with the Threads API. When is_made_with_ai=true is sent while creating a media container, an AI info label appears on the published post; for carousels, the parameter must be used on the parent container.
Usage details
Who can use it: Apps using the Threads API with the threads_content_publish permission and required access; text, image, video, and carousel posts.
How to use it: Add is_made_with_ai=true when creating a media container in a POST /{threads-user-id}/threads request. For carousels, use the parameter on the parent container with media_type=CAROUSEL; then publish the creation_id through the existing POST /{threads-user-id}/threads_publish flow.
Tip: Clearly label AI-assisted creations; use the parameter on the parent container rather than setting it to true on carousel child containers.
✓ Verified by official sourceSource: Threads API changelog, 30 Ekim 2026Source: Threads gönderileri: Label AI-Generated ContentA shared AI training preference signal to be supported for websites
OpenAI announced that it will follow a shared signal that lets websites use robots.txt to tell multiple AI providers at once whether their content may be used to train foundation models. Implementation details will be shared in the coming months; existing OAI-SearchBot and GPTBot controls remain in effect.
Usage details
Who can use it: For publishers and website owners; OpenAI’s implementation of the new signal is not live yet and is planned for the coming months.
How to use it: For now, continue managing OAI-SearchBot and GPTBot rules separately in robots.txt; there is no step to enable the new shared signal until OpenAI publishes the implementation syntax.
Tip: Don’t confuse visibility in ChatGPT search with preferences for training use; maintain separate rules for OAI-SearchBot and GPTBot.
/design-sync upload confirmation shows more detail about files to be sent
The /design-sync upload prompt now clearly shows the folder, project, and files to be uploaded or deleted.
Usage details
Who can use it: The /design-sync workflow in Claude Code.
How to use it: Run the /design-sync command; review the folder, project, and file list in the confirmation prompt.
Tip: Check the list of files to be deleted especially carefully before confirming.
✓ Verified by official sourceSource: Claude Code v2.1.314 GitHub sürüm notlarıAutoClaw offers rewards of up to 80M tokens in its Halloween campaign
A new 40M-token reward (4,000 credits) can be claimed during the October 31–November 8 UTC+8 period. If the previous week's reward has not yet been claimed, the total can reach up to 80M tokens across both rewards; the older weekly reward expires on November 1 at 23:59 UTC+8.
Usage details
Who can use it: AutoClaw users can claim the new reward starting October 31. The new campaign runs through November 8 at 23:59 UTC+8; the previous reward has a separate expiration date.
How to use it: Log in to AutoClaw starting October 31 UTC+8 and claim the new 40M-token reward. If you also claimed the previous week's 40M reward, use it before November 1 at 23:59 UTC+8; the two rewards may not be combined into a single balance.
Tip: Track the expiration dates of the two rewards separately: November 1 for the old credit and November 8 for the new campaign.
✓ Verified by official sourceSource: Halloween sürprizi: 80M tokene kadar talep edinThe Bash tool no longer passes inherited which functions to commands
The Bash tool now cleans up inherited which alias or function definitions and doesn’t pass exported which or unsetenv functions to commands, so the system command or zsh builtin is used.
Usage details
Who can use it: Claude Code Bash tool; especially shell environments that use aliases or functions.
How to use it: Run which with the Bash tool without any extra setup; the cleaned-up shell behavior is applied automatically.
Tip: Test scripts that rely on custom which behavior in your environment against the system command behavior.
✓ Verified by official sourceSource: Claude Code v2.1.314 GitHub sürüm notlarıguardrail_trace added to Bedrock upstreams
The guardrail_trace option has been added to Claude apps gateway Bedrock upstreams. It records which Amazon Bedrock guardrail filter was triggered during a request, but does not log the matching text.
Usage details
Who can use it: Claude apps gateway administrators using an Amazon Bedrock upstream.
How to use it: Add the guardrail_trace option to the relevant Bedrock upstream configuration and review the resulting traces in your gateway monitoring/logging system.
Tip: Track filter names for security diagnostics; since sensitive matching text is not recorded, correlate the incident’s underlying content with separate access-controlled logs.
✓ Verified by official sourceSource: Claude Code v2.1.314 sürüm notlarıThe CFC prompt variable also stops Chrome instructions from being added to VS Code messages
CLAUDE_CODE_DISABLE_CFC_PROMPT now also stops the VS Code extension from adding Claude in Chrome instructions to session messages. Browser tools remain available.
Usage details
Who can use it: Claude Code VS Code extension; browser tools remain available.
How to use it: Set the CLAUDE_CODE_DISABLE_CFC_PROMPT environment variable before starting the VS Code extension.
Tip: Use only this variable if you want to keep using browser tools while removing the automatic Chrome instructions.
✓ Verified by official sourceSource: Claude Code v2.1.314 GitHub sürüm notlarıClaude Code’s built-in API and dataviz skills updated
The Datadog API monitoring setup in the built-in /claude-api skill now sends the dd-llmobs-enabled: true header and removes the old macOS xattr step. The built-in dataviz skill also recommends printing axis labels more legibly and using sorted bars or segmented bars for pie/donut requests.
Usage details
Who can use it: Users of Claude Code’s built-in Skills package, as well as teams using the /claude-api skill.
How to use it: Use the /claude-api or dataviz skill in a Claude Code session; the updates come with the built-in package, so no separate installation is needed.
Tip: When checking Datadog traces, verify that the dd-llmobs-enabled header was received; prefer the new bar formats for category comparisons in charts.
✓ Verified by official sourceSource: Claude Code v2.1.314 sürüm notlarıCloud sessions started from the CLI get their session and branch names from the server
Cloud sessions started from the CLI now get their session and branch names from the server instead of the local client.
Usage details
Who can use it: Cloud sessions started from the Claude Code CLI.
How to use it: Start a Cloud session from the CLI; the session and branch names will appear with the values returned by the server.
Tip: After renaming the local branch, check that its name in the Cloud session matches the server.
✓ Verified by official sourceSource: Claude Code v2.1.314 GitHub sürüm notlarıRepository and environment lists are displayed as tables in Cloud session transcripts
Repository and environment lists in Cloud session transcripts are now displayed as readable tables instead of single-line JSON; the rename row also clearly identifies the new title.
Usage details
Who can use it: Claude Code Cloud sessions.
How to use it: Open a Cloud session and view the transcript; repository and environment details appear in table format.
Tip: Use the table columns to compare long environment names; there’s no need to copy raw JSON.
✓ Verified by official sourceSource: Claude Code v2.1.314 GitHub sürüm notlarıThe repository-addition step for Cloud sessions now shows a summary table
When Claude adds a repository to a Cloud session, opening the relevant step now shows a brief table of actions performed instead of raw JSON.
Usage details
Who can use it: Claude Code Cloud sessions.
How to use it: Add a repository when creating a Cloud session; open the repository-addition step and review the summary table.
Tip: When adding multiple repositories, check the table to verify the branch and repository pairings.
✓ Verified by official sourceSource: Claude Code v2.1.314 GitHub sürüm notlarıCtrl+R history search now finds text in collapsed pasted content
When searching prompt history with Ctrl+R, text inside collapsed [Pasted text] sections is now included in the search.
Usage details
Who can use it: Claude Code terminal interface.
How to use it: In a terminal session, press Ctrl+R and search for a phrase from previously pasted text.
Tip: When reusing long logs or code blocks, search for a distinctive word.
✓ Verified by official sourceSource: Claude Code v2.1.314 GitHub sürüm notlarıDeepSeek Harness rc.5 improves how thinking content is displayed in conversations
DeepSeek Harness v0.2.1-rc.5 released a UI update that improves how thinking content is displayed in conversations. The official release notes do not mention any other features or settings for this release.
Usage details
Who can use it: Users who upgrade to the DeepSeek Harness v0.2.1-rc.5 prerelease; conversation view.
How to use it: Upgrade DeepSeek Harness to v0.2.1-rc.5 or later and open a conversation; the improvement noted in the release notes applies to how thinking content appears in the conversation. The release notes do not mention any additional setting or command that needs to be enabled.
Tip: When reviewing long thinking responses, start with the new display in this release; when needed for debugging, save the response's thinking sections together with the conversation context.
✓ Verified by official sourceSource: GitHub resmî sürüm notu: DeepSeek Harness v0.2.1-rc.5Source: Freedom.Tech: DeepSeek Harness 0.2.1-rc.5FORCE_HYPERLINK environment variable fixed to disable links
In addition to FORCE_HYPERLINK=false, the values no and off now disable terminal links; the previous inverse behavior has been changed.
Usage details
Who can use it: Claude Code terminal output.
How to use it: Set FORCE_HYPERLINK=false, FORCE_HYPERLINK=no, or FORCE_HYPERLINK=off to disable links.
Tip: Explicitly set this variable to prevent link escape sequences in CI or plain-text logs.
✓ Verified by official sourceSource: Claude Code v2.1.314 GitHub sürüm notlarıClaude Code gives better guidance when no GitHub remote is found
If /ultrareview, cloud session, or cloud agent cannot find a GitHub remote, the message now suggests creating a repository via github.com/new; /ultrareview also no longer suggests an invalid gh command.
Usage details
Who can use it: Claude Code /ultrareview, cloud session, and cloud agent workflows.
How to use it: Start the relevant workflow in a workspace without a remote; follow the notification’s suggested path to create a GitHub repository.
Tip: Adding a remote before review ensures findings are written back to the correct repository.
✓ Verified by official sourceSource: Claude Code v2.1.316 GitHub sürüm notlarıBuilt-in prompts optimized in Kimi Code CLI and Desktop
The October 30 release notes for Kimi Code CLI 2.4.1 and Kimi Code Desktop 1.2.1 say that the built-in prompts have been optimized. The official notes do not explain the scope or measurement of the optimization.
Usage details
Who can use it: Kimi Code CLI 2.4.1 and Kimi Code Desktop 1.2.1 users; no platform or plan restrictions specified.
How to use it: Update Kimi Code CLI to 2.4.1 or Kimi Code Desktop to 1.2.1; no additional setting or command is required for this change.
Tip: Since Moonshot has not published any measurements, record the outputs of a few repeatable tasks before updating and compare the behavior afterward.
✓ Verified by official sourceSource: Kimi Code CLI değişiklik günlüğüSource: Kimi Code Desktop değişiklik günlüğüKimi Code Desktop reduces context usage for long Chinese files
Kimi Code Desktop 1.2.1 reduces the amount of context used when reading long Chinese files. Built-in prompts were also optimized in the same release.
Usage details
Who can use it: Desktop users running Kimi Code Desktop 1.2.1 or later; the release notes do not specify an operating system.
How to use it: Update Kimi Code Desktop to 1.2.1; the improvement is applied automatically when reading long Chinese files and requires no additional settings.
Tip: For very long files, attaching only the relevant sections or files to the task can still make context usage more predictable.
✓ Verified by official sourceSource: Kimi Code Desktop değişiklik kaydı — 1.2.1 (30 Ekim 2026)Confirmation required when editing the organization settings file
Claude Code’s file tools now ask for confirmation before editing a local copy of the organization’s server-managed settings. This behavior is now consistent with the protection for users’ own settings files.
Usage details
Who can use it: Claude Code users whose organization applies server-managed settings.
How to use it: When Claude Code asks to write to the organization settings file, review the permission prompt and approve only the expected change.
Tip: When possible, update the policy in its management source instead of directly changing the managed settings file.
✓ Verified by official sourceSource: Claude Code v2.1.314 sürüm notlarıSession titles can now be generated automatically from skill- or prompt-based commands
If a session starts with a skill or a prompt-based command such as /security-review, Claude Code now generates a title automatically from that starting point, so the session is no longer left untitled.
Usage details
Who can use it: Claude Code terminal and related session lists.
How to use it: Start a new session with a skill or prompt-based slash command; a title will be created automatically.
Tip: Choose short, descriptive starting commands to make sessions easier to find later.
✓ Verified by official sourceSource: Claude Code v2.1.314 GitHub sürüm notlarıInclude completed button added to the Scheduled tasks page
On the Scheduled tasks page in Cloud sessions, an Include completed button appears only when completed one-time tasks are hidden. The button makes completed tasks visible without opening the Sort menu.
Usage details
Who can use it: Users of the Scheduled tasks view in Claude Code cloud sessions.
How to use it: Open the Scheduled tasks page; if completed one-time tasks are hidden, click the Include completed button.
Tip: Use this button to review one-time task history; continue using the existing sort options to filter recurring tasks.
✓ Verified by official sourceSource: Claude Code v2.1.314 sürüm notlarıSSL certificate errors now explain the likely cause and how to fix it
For SSL certificate errors, Claude Code now identifies likely causes, such as a missing intermediate certificate, and recommends a fix for each cause.
Usage details
Who can use it: Claude Code network connections.
How to use it: If you get an SSL error, follow the cause and resolution steps in the new error message.
Tip: First check that the server is actually sending the intermediate certificate; don’t change the client certificate store at random.
✓ Verified by official sourceSource: Claude Code v2.1.316 GitHub sürüm notlarıThe VS Code usage limit reset is now shown as a time
The usage limit card in the VS Code extension now shows a time such as “Resets at 3:10 PM” instead of a generic reset message, and labels the window as 5-hour limit.
Usage details
Who can use it: Claude Code VS Code extension.
How to use it: When you reach the usage limit, open the limit card in VS Code to see the reset time directly.
Tip: Break long tasks into parts before the reset time.
✓ Verified by official sourceSource: Claude Code v2.1.314 GitHub sürüm notlarıFiles can be created and edited in Gemini chats
Gemini can now create, preview, and edit documents, PDFs, presentations, spreadsheets, Markdown, and code files in chat. PDF editing is supported on web and mobile; other formats are supported on web, and results can be exported to Google Slides, Sheets, or Drive.
Usage details
Who can use it: Personal accounts with access to the Gemini app, Workspace Individual, and Google Workspace users; in all supported languages.
How to use it: Ask Gemini to create the file type you want in a chat; open the preview full-screen, select text, and request edits.
Tip: Check the structure and sources in the Gemini preview before exporting a presentation or spreadsheet.
✓ Verified by official sourceSource: Create, preview, and edit files directly in the Gemini appJules will be discontinued on December 31 and become read-only on November 11
Google is discontinuing Jules to consolidate its agentic coding work in Antigravity. Users can finish their sessions through November 11, 2026; after that date, Jules will be read-only, and the service will shut down completely on December 31, 2026. Settings, environment variables, repositories, memories, scheduled tasks, and session history stored in Jules can be exported to a local folder.
Usage details
Who can use it: All Jules users are affected. The deadline to finish new sessions is November 11, 2026; from November 11 through December 31, access will be read-only, and Jules will be unavailable after December 31. Code, branches, and pull requests on GitHub will remain in place.
How to use it: To export your data, install Jules Tools in a terminal with `npm install -g @google/jules` and sign in with `jules login`. Then run the export prompt from the official announcement in a CLI coding agent; the process saves settings, environment variables, connected repositories, memories, setup scripts, and all session history to a local `jules-export` folder.
Tip: Finish ongoing sessions and export your data before November 11. Keep the export folder private, since `env-variables.json` may contain secrets.
Work and Codex usage doubled on Pro and Business Premium plans
Work and Codex usage allowances on Pro 100 and Pro 500 plans have been doubled from their previous levels. The previously announced Pro 200 reduction was also canceled, and subscribers with lower allowances were raised to the same level.
Usage details
Who can use it: ChatGPT Pro 100, Pro 200, and Pro 500, and Business Premium; depends on the plan and account status.
How to use it: No action is required; the new usage allowance is applied to your account automatically.
Tip: Instead of saving intensive Work or Codex tasks for the end of the month, use the new allowance for longer-running tasks.
Usage and Cost Analytics API launches in beta
The new Usage and Cost Analytics API offers two beta endpoints for reporting token usage and costs for Claude Console and Claude Enterprise organizations. Data can be filtered and grouped by dimensions such as model, workspace, API key, and user or service account.
Usage details
Who can use it: Beta for Claude Console and Claude Enterprise organizations; the analytics-usage-cost-2026-08-24 beta header is required.
How to use it: Send POST requests to the /v1/organizations/analytics/usage and /v1/organizations/analytics/cost endpoints with the relevant beta header.
Tip: Track costs by workspace and service account as well as by model to identify budget variances.
✓ Verified by official sourceSource: Claude Platform API sürüm notlarıAccess to and training on AI glasses expanded in Canada through CNIB partnership
Meta launched a program with CNIB to help blind, deafblind, or partially sighted people in Canada try AI glasses at work and in daily life. Through CNIB’s Come to Work program, the initiative will provide donated glasses, peer training, accessible guides, and feedback to Meta’s product teams.
Usage details
Who can use it: The program is aimed at blind, deafblind, or partially sighted people in Canada, especially those participating in the CNIB Come to Work program. The announcement describes targeted access and training through CNIB, not an automatic nationwide rollout or sale for general users.
How to use it: This is not a feature enabled through a general Meta AI setting. Interested people should apply to the training and demonstration process through the CNIB Come to Work program; eligible glasses should be paired with the Meta AI mobile app. The announcement does not provide a separate self-service activation menu or a public application URL.
Tip: During program demonstrations, try existing AI glasses use cases—such as reading text, recognizing objects, and describing surroundings—with real work tasks. Use the accessible guides and feedback channel to tell CNIB which tasks are not adequately supported.
✓ Verified by official sourceSource: Meta ve CNIB, Kanada’da AI gözlüklerine erişimi destekliyorCursor and live macOS PiP enabled by default in CUA
Qwen Code v0.26.4 enables default cursor behavior and a live picture-in-picture view on macOS in computer-use (CUA) workflows. This makes it easier to monitor the model’s screen interactions and see the active target.
Usage details
Who can use it: Available in macOS CUA workflows using Qwen Code v0.26.4; the release note specifically identifies live PiP support as a macOS feature.
How to use it: Update Qwen Code to v0.26.4 and start a CUA session on macOS; `default cursor` and `live macOS PiP` behavior are enabled with the release.
Tip: Before starting a CUA task, verify the active app and target area in the PiP window; visually check actions that are difficult to reverse, especially deleting or sending files.
✓ Verified by official sourceSource: Qwen Code v0.26.4 sürüm notlarıLegacy Usage and Cost reports are now legacy
The usage_report, cost_report, user_usage_report, and user_cost_report endpoints in the Usage and Cost API and Enterprise Analytics API have been marked legacy. Existing integrations will continue to work, but migration to the Analytics API is recommended for new development.
Usage details
Who can use it: Support continues for existing users; the new Analytics API is recommended for new integrations.
How to use it: Follow the Migrate from the Usage and Cost API guide to move to the new usage and cost endpoints.
Tip: Instead of removing the old endpoints immediately, compare their results with the new API for a while during migration.
✓ Verified by official sourceSource: Claude Platform API sürüm notlarıGenerated UI cards now use theme-compatible interactive views
Qwen Code v0.26.4 presents Generated UI views in card frames that follow the host theme, can be localized, and can be reshaped. Cards can open as interactive views when users want to adjust values, marking the next step for the previous editable-card capability in the stable release.
Usage details
Who can use it: Available in the Generated UI host/WebShell surfaces of Qwen Code v0.26.4; some newer default Generated UI behaviors are available only in the October 29 nightly preview.
How to use it: Update to v0.26.4 and create a Generated UI card in WebShell. Open the interactive view inside the card to edit values directly; the release applies the host theme and card frame.
Tip: Adjusting parameters in the card and sending a follow-up question through the same card is faster than writing a new long prompt for every change.
✓ Verified by official sourceSource: Qwen Code v0.26.4 sürüm notlarıGLM-5.3-FlashX-Legal is now available directly in AutoClaw
GLM-5.3-FlashX-Legal has entered the stage where GLM-Legal can be used directly in AutoClaw. The model can review contracts, cross-check evidence, and conduct compliance research; with AutoClaw's agent runtime and document tools, it can produce tracked changes and comments in Word files.
Usage details
Who can use it: Available directly to AutoClaw users. Access to the model API under the `glm-5.3-flashx-legal` identifier is open to selected enterprise customers by application; specialized MCP services for legal work and AutoClaw Legal Assistant are also recommended.
How to use it: Open AutoClaw and provide your legal task and source documents. If using the API, use `glm-5.3-flashx-legal` as the model parameter and apply for selected enterprise trial access at https://z.ai/consultation/legal.
Tip: Provide source files together for contract and evidence tasks; enable tracked changes and comments for document revisions, and have a legal professional approve the final text.
✓ Verified by official sourceModels: GLM-5.3-FlashX-Legal, GLM-Legal, GLM-5.3-FlashSource: GLM-5.3-flashX-Legal AutoClaw'da kullanıma açıldıSource: GLM-5.3-Flash/FlashX ve GLM-5.3-FlashX-Legal API kılavuzuGPT-6 Luna gives shorter, less sycophantic responses
OpenAI updated GPT-6 Luna in ChatGPT to provide shorter responses and reduce overly affirming language. The update aims for more direct and balanced results in everyday conversations.
Usage details
Who can use it: In the ChatGPT Chat experience, depending on account and phased rollout status.
How to use it: In ChatGPT, use the GPT-6-based Chat experience from the model selector; no additional setting needs to be enabled.
Tip: To make responses even shorter, specify the target length and format clearly in your prompt.
✓ Verified by official sourceModels: GPT-6 LunaSource: GPT-6 Luna 29 Ekim güncellemesi güvenlik ek notuGrok Bot 0.78.1 begins stable rollout: shopping and subscription templates
Grok Bot 0.78.1 began rolling out gradually on the stable channel following an announcement dated October 28, 2026, at 23:27:36 UTC; this corresponds to October 29 in Istanbul. The release shows product results as cards/carousels with photos, ratings, prices, and store information; users can make approved purchases with Buy now and track deliveries. It also adds templates for subscription trials, finding unused subscriptions, and tracking renewal prices.
Usage details
Who can use it: The stable release is rolling out to accounts with Grok Bot access; xAI has not published a separate country list. Access depends on Grok Bot plans such as The X Subscription Premium/Super/Ultra, eligible legacy Cursor plans, and linked SuperGrok or X Premium+ accounts; the feature may appear gradually depending on the account’s rollout and the desktop or mobile Grok Bot app you use.
How to use it: Update Grok Bot and ask a Bot to search for a product; open a product from the results card, select Buy now, and approve the confirmation card. Connect Shop Pay through Settings → Vault → Payments; then select it in the Pay with field or make it the Bot’s first payment method. For subscription templates, ask a Bot which templates can help with subscriptions, then add one of the Try it, don’t pay for it, Find unused subscriptions, or Watch subscription prices templates.
Tip: Set purchases and subscription cancellations to Ask first in Auto Review. If you want separate approval for every payment, don’t save the card permanently to Vault; a saved card may be filled in during payment without additional confirmation.
Kimi Work asks users for critical information before a task
With Kimi Work 3.2.22, Agent gained the Ask user feature, which confirms critical information with the user before running a task. This lets missing or ambiguous inputs be clarified before the task starts.
Usage details
Who can use it: Users running Kimi Work desktop app version 3.2.22 or later.
How to use it: Update Kimi Work to 3.2.22 and start a task; review the Ask user confirmation that appears before execution and provide the necessary information.
Tip: For tasks that affect external systems or take a long time, carefully review the assumptions and scope in the confirmation screen before running them.
✓ Verified by official sourceSource: Kimi Work sürüm notları — 3.2.22 (29 Ekim 2026)Manually purchase an Extra Usage pack after a quota warning
Extra Usage packs became available with Kimi Work 3.2.22; subscribers can purchase additional usage balance when they receive a quota warning. The balance is shared across Kimi on web, browser extension, desktop, and mobile, as well as Kimi Code.
Usage details
Who can use it: Subscribers only; available for personal accounts on web, in the browser extension, desktop and mobile apps, and in Kimi Code. Enterprise plans are not currently supported.
How to use it: On the web, open the Extra Usage Credit Pack section on Kimi’s Subscription page and select an amount. In Kimi Code, use the Buy Extra Usage Pack button that appears when you exceed 80% of the rolling 5-hour usage limit or the weekly limit on the old plan; then activate Extra Usage Credit Pack on the web.
Tip: To limit unexpected spending, open Settings in the Extra Usage section on the web and enable Monthly spending cap.
Qwen Code can install and update unscoped npm extensions
With Qwen Code v0.26.4, npm extensions without a scope can be installed and updated through the CLI and daemon. The source can be specified as a package name, a specific version, or a tag; updates retain the registered package source and registry.
Usage details
Who can use it: Applies to CLI and daemon users running the stable Qwen Code v0.26.4 release; scoped and unscoped extensions published to the npm registry are supported.
How to use it: In the extension installation or update workflow, use `package-name`, `package-name@version`, or `package-name@tag` as the source. The same npm source formats apply to daemon installations.
Tip: For reproducibility in production, pin a version with `package-name@version`; use the `@latest` and `@beta` tags only in controlled update workflows.
✓ Verified by official sourceSource: Qwen Code v0.26.4 sürüm notlarıSource: Qwen Code PR #15781: scope’suz npm kurulumu ve güncellemeleriQwen Code expands WebShell workspace and session management
v0.26.4 displays sessions without workspaces alongside workspaces in WebShell, lets embedded hosts select a workspace directory, and adds independent worktrees for Managed parent sessions. The same release also raises the workspace skill installation limit.
Usage details
Who can use it: Applies to WebShell, Managed sessions, and host applications embedding Qwen Code when using Qwen Code v0.26.4.
How to use it: Open the `Session Overview` view in WebShell to manage sessions without workspaces; use the `workspace directory` selection in embedded host integrations. Choose the `independent worktree` option for Managed parent sessions.
Tip: Using independent worktrees while handling multiple tasks or branches at once helps prevent file changes from getting mixed together.
✓ Verified by official sourceSource: Qwen Code v0.26.4 sürüm notlarıScheduled section consolidated as Task Center
In Kimi Work 3.2.22, Scheduled was renamed Task Center. Conversation tasks, widget tasks, and other automation tasks can be viewed and managed in one place.
Usage details
Who can use it: Users running Kimi Work desktop app version 3.2.22 or later; the relevant task feature is also available in Kimi.
How to use it: Open Task Center in Kimi Work; enable or pause tasks, run them once, edit them, or delete them. Conversation and widget tasks are also listed in the same center.
Tip: Before enabling a new automation, test its output using Run once now on the task card.
✓ Verified by official sourceSource: Kimi Work sürüm notları — 3.2.22 (29 Ekim 2026)Source: Kimi görevleri yardım sayfasıAdd custom DIY skins to the Kimi Work desktop pet
Kimi Work 3.2.22 lets users customize the desktop pet’s appearance and add their own DIY skins.
Usage details
Who can use it: Users running Kimi Work desktop app version 3.2.22 or later.
How to use it: Update the app to 3.2.22; add your own DIY skin from the desktop pet customization area. The release notes do not specify the exact menu path.
Tip: Try a small, simple skin first; if display issues occur, switch back to the default skin.
✓ Verified by official sourceSource: Kimi Work sürüm notları — 3.2.22 (29 Ekim 2026)Quoted reply setting now configurable by conversation type
Qwen Code v0.26.4 supports configuring quoted replies by conversation type in DWS workflows. This lets you manage separately whether to quote a previous message in each conversation format.
Usage details
Who can use it: Available in bot and conversation workflows that use DWS integrations in Qwen Code v0.26.4.
How to use it: Set the `quoted replies` behavior for `conversation type` in the DWS conversation configuration; enable quoting for the conversation types that need it and disable it for others.
Tip: Enable quoting in support or review conversations where it is critical to identify which message a reply refers to; disable it in quick chats to reduce unnecessary context.
✓ Verified by official sourceSource: Qwen Code v0.26.4 sürüm notlarıPhased upgrade command added to Managed Agent
Qwen Code v0.26.4 adds a phased upgrade command for Managed Agent in the reference deployment. Upgrades can now be carried out through defined stages instead of in a single step.
Usage details
Who can use it: Available to developers and administrators using Qwen Code’s v0.26.4 reference Managed Agent deployment.
How to use it: Upgrade the reference deployment to v0.26.4 and run the `phased upgrade` command in the Managed Agent management workflow; do not proceed to the next stage until the current stage is complete.
Tip: Before upgrading production, run the same phases in staging and verify session and runtime state at each stage.
✓ Verified by official sourceSource: Qwen Code v0.26.4 sürüm notlarıThe Model Spec has been reorganized around five core values
OpenAI has reorganized its Model Spec guidance around the values of faithfulness to human intent, minimizing harm, honesty and transparency, human autonomy, and care. New guidance on context, oversight, and autonomy has been added; most detailed style guidance has been removed, and OpenAI notes that the document describes intended behavior that models do not yet fully reflect.
Usage details
Who can use it: Applies to the public Model Spec and intended model behavior; there is no specific plan, region, or on/off setting, and full behavioral rollout is not guaranteed.
How to use it: Review the updated document at https://model-spec.openai.com/2026-10-29.html; there is no separate setting in ChatGPT to enable it.
Tip: State your preferred tone, perspective, and boundaries explicitly; many of the guide’s defaults can be steered with user or developer instructions, but you should not expect them to be implemented fully in every model.
Generated UI becomes the default and WebShell workflows expand in nightly release
The October 29 preview, `v0.26.4-nightly.20261029.70be140e14`, makes Generated UI the default and lets users continue conversations using card values. The same nightly release adds Session Overview and Split View keyboard shortcuts, a background agent counter, live message pagination in WebShell, one-click update coverage, and Russian localization for the Goal card.
Usage details
Who can use it: Available only in the Qwen Code nightly/preview channel; it is not automatically available to stable v0.26.4 users.
How to use it: Install the exact `v0.26.4-nightly.20261029.70be140e14` asset from the GitHub Releases page. In WebShell, use `Session Overview`, `Split View`, the Goal card, and background agent status; if the one-click option is unavailable on the update screen, follow the manual steps shown.
Tip: Try the nightly build in a separate QWEN_HOME or test environment rather than your stable installation; verify card interactions before bringing the Generated UI default into production sessions.
✓ Verified by official sourceSource: Qwen Code v0.26.4 nightly sürüm notlarıQoder extends the DeepSeek-Flash free usage period through October 31
Qoder International has extended its daily free usage promotion for individual users of DeepSeek-Flash (DeepSeek-V4.1-Flash-1017) through the end of October 31, 2026, at 24:00 (UTC+8). Usage within the daily free quota consumes 0.0× credits; the standard credit rate applies when the quota or promotion period ends.
Usage details
Who can use it: Individual accounts on Qoder International’s Free, Pro Trial, Lite, Pro, Pro+, and Ultra plans are eligible; Teams and Enterprise accounts are excluded. The promotion is available in Qoder, Qoder IDE, the JetBrains plugin, CLI, Cloud Agents, Mobile, and the web app.
How to use it: In the Qoder product you use, select `DeepSeek-Flash` from the model picker and start a task. The promotion is applied automatically; no application or additional settings are required, and eligible individual accounts with a zero credit balance can also use the daily quota.
Tip: Plan your use of the daily free quota before it expires; maximum reasoning effort also consumes 0.0× credits within the quota, but the standard credit rate applies once the quota is exceeded and after November 1.
Model and tool versions are stamped on verify-pr reports
Qwen Code v0.26.4 adds the versions of the model, CLI, and skill used to published verify-pr reports. This makes it easier to reproduce the report with the same toolchain and audit it later.
Usage details
Who can use it: Available to developers using the verify-pr reporting flow in Qwen Code v0.26.4.
How to use it: Update to v0.26.4 and publish a verify-pr report; the model, CLI, and skill version fields are included automatically in the report output.
Tip: Keep these version fields unchanged when sharing the report; do not treat results obtained with a different model or skill version as the same review.
✓ Verified by official sourceSource: Qwen Code v0.26.4 sürüm notlarıModel timings added to WebShell Token Usage panel
Qwen Code v0.26.4 now displays model timing inline in the Token Usage view in WebShell. Users can monitor the timing of a model call in the same place where they view token consumption.
Usage details
Who can use it: Available to Qwen Code v0.26.4 WebShell users.
How to use it: Open the `Token Usage` panel in a WebShell session; model timing information appears inline in the usage view.
Tip: When comparing the same prompt across different models, track model timing alongside token counts to distinguish whether latency comes from the model or output length.
✓ Verified by official sourceSource: Qwen Code v0.26.4 sürüm notlarıImporting Claude Code and Codex sessions added
Kimi Code CLI 2.4.0 added a feature that converts Claude Code and Codex sessions from the past 15 or 30 days into regular Kimi Code sessions, letting you pick up where you left off.
Usage details
Who can use it: Kimi Code CLI 2.4.0 users.
How to use it: Run the /import-sessions command or the kimi import-sessions command, then select the time range.
Tip: Back up the old tool’s session folders before migrating; import only the date range you need.
✓ Verified by official sourceSource: Kimi Code CLI ChangelogTexting with Dot via iMessage or RCS begins
The feature, which is starting as a beta for Pro users in the US, lets users reach dot via iMessage or RCS. Dot can also send users proactive updates; memory and plugin context are shared with dot in ChatGPT.
Usage details
Who can use it: Phased beta rollout to Pro users in the US.
How to use it: Open the dot profile in ChatGPT and select Texting; enable iMessage on iPhone or RCS on Android.
Tip: Review your dot instructions and connected plugins to limit proactive updates sent by message.
✓ Verified by official sourceSource: ChatGPT yayın notlarıGemini 4 Carbon announced
Google announced Gemini 4 Carbon, focused on long-running coding and agent workflows, as well as cybersecurity defense. The model is initially available to a limited group of cyber defenders through the Fairwind Program; general access for developers, businesses, and consumers is planned for later.
Usage details
Who can use it: For now, selected cyber defenders and Fairwind Program partners; general user and developer access is not yet available.
How to use it: Once general access is available, use the Gemini 4 Carbon model selection in the Gemini API, product surfaces, or developer consoles.
Tip: For cybersecurity use, add independent security testing and human approval before automatically applying model outputs.
✓ Verified by official sourceModels: Gemini 4 CarbonSource: Gemini 4 Carbon: our next frontier modelPlan mode now starts only when enabled by the user
With CLI 2.4.0, Agent can no longer switch to Plan mode on its own; every plan awaits user review, and the Never Ask setting does not automatically approve plans.
Usage details
Who can use it: Kimi Code CLI 2.4.0 users.
How to use it: Enable Plan mode yourself using the plan setting in the session or the relevant command, then approve the suggestion.
Tip: If you want to review a plan before files are touched, this change provides a safer review workflow.
✓ Verified by official sourceSource: Kimi Code CLI ChangelogSmart reports, surveys, and pooled group budgets reach GA
Smart reports, in-app surveys, and pooled group budgets are now generally available in Claude Enterprise. Administrators can see what teams are working on and spending, collect feedback, and share a monthly budget across teams.
Usage details
Who can use it: Claude Enterprise organizations; eligibility may vary with configurations such as HIPAA or customer-managed encryption.
How to use it: Enable Smart reports, surveys, and pooled group budgets in admin settings; select a team and date range to create a report.
Tip: Use reports to identify repeatable skills and missing connectors, not to measure individual performance.
✓ Verified by official sourceSource: Claude Enterprise Smart reports duyurusuModel and concurrency selection expanded for Tower tasks
CLI 2.4.0 added support for per-task model selection and the max_running_tasks limit in Tower mode; the default Agent concurrency limit also increased from 16 to 256.
Usage details
Who can use it: Members using Kimi Code CLI 2.4.0 and Tower.
How to use it: Set the task model and max_running_tasks value in the Tower configuration.
Tip: When running many tasks at once, start with a low max_running_tasks value to control costs and quotas.
✓ Verified by official sourceSource: Kimi Code CLI ChangelogMonthly promotional API credits for eligible Pro and Business subscribers
Eligible ChatGPT Pro and Business subscribers can activate monthly promotional API credits to develop apps and agents. The credits are separate from ChatGPT usage and are designed to be used in the connected API organization.
Usage details
Who can use it: Eligible ChatGPT Pro and Business subscribers; credit amount and eligibility vary by account and offer.
How to use it: Connect your own API organization in ChatGPT or the API Platform and activate the subscription benefit.
Tip: Use the credits first for low-risk prototypes, and check the API organization's billing settings separately.
Setting added to prevent the computer from sleeping
Kimi Code Desktop 1.2.0 introduced a setting that prevents the computer from sleeping during long-running Agent tasks.
Usage details
Who can use it: Kimi Code Desktop 1.2.0 users.
How to use it: Enable the sleep prevention setting under Settings → General.
Tip: Turn it on when running a long build or background task, then turn it off when the work is done to reduce battery consumption.
✓ Verified by official sourceSource: Kimi Code Desktop ChangelogFill with Gemini now shows sources at the cell level
In Google Sheets, Fill with Gemini now shows the web sources it uses separately in each cell. Request capacity has increased to 3,000 cells, while processing speed and consistency across cells have also improved.
Usage details
Who can use it: Users with Gemini in Sheets enabled on Business Standard/Plus, Enterprise Standard/Plus, Google AI Pro/Ultra, Google AI Pro for Education, and AI Expanded Access.
How to use it: In Google Sheets, open the Ask Gemini icon in the upper right and fill cells with AI function; sources appear in the cells.
Tip: Before using generated data, verify the source links in cells by sampling them.
✓ Verified by official sourceSource: Fill with Gemini now provides cell-level citations for web sources and supports larger requestsReady-to-use Artifacts gallery for legal teams
Claude for Legal now includes an adaptable gallery of ready-to-use Artifacts for tasks such as contract review, regulatory research, cite checks, and clause comparison. Teams can add their own documents and connectors, then share outputs with source links.
Usage details
Who can use it: Claude for Legal and Claude Enterprise teams; subject to organization configuration and relevant connectors.
How to use it: Open the Artifacts gallery in Claude or the Claude for Legal page, then choose a legal template.
Tip: Use the automated initial review as a source-linked preliminary step, not as a substitute for a lawyer’s assessment.
Layered interactive explorations roll out
The Gemini app has added an interactive exploration library that lets users examine topics layer by layer, from microscopic systems to space. Users can tap nodes to explore further with guided explanations; audio narration is also planned.
Usage details
Who can use it: All users on web and mobile; the full experience is available in English, Indonesian, Hindi, Brazilian Portuguese, French, and Simplified Chinese. In other languages, the experience is in English.
How to use it: Open the Students section from the Gemini menu or go to gemini.google.com/immersive-explorations.
Tip: When studying a topic, move from large-scale concepts to atomic details to build a concept map step by step.
✓ Verified by official sourceSource: Immersive explorations now available in the Gemini appMessages and Token Counting request limit raised to 128 MB
The maximum request size for the Messages API and Token Counting API in the first-party Claude API has increased from 32 MB to 128 MB. This reduces the need to split up workflows that use large files and multiple content items.
Usage details
Who can use it: First-party Claude API.
How to use it: Send the request body directly, ensuring it does not exceed the 128 MB limit; also check your client and gateway limits.
Tip: For large requests, still count tokens in advance and test timeout and proxy limits.
✓ Verified by official sourceSource: Claude Platform API sürüm notlarıAutomatic mention support added for pasted images
Desktop 1.2.0 automatically adds pasted images to the Agent context as mentions.
Usage details
Who can use it: Kimi Code Desktop 1.2.0 users.
How to use it: Turn on the automatic image mention setting under Settings → Agent & Sessions → Composer.
Tip: Keep this setting on so you don’t have to select an image separately when requesting edits based on a screenshot.
✓ Verified by official sourceSource: Kimi Code Desktop ChangelogCustom moderation guidelines in API requests
The `guidelines` array added to Grok API requests lets you define `decline` rules that make the model stricter in the nsfw, violence, and profanity categories. This allows developers to set application-specific safety boundaries in addition to default moderation.
Usage details
Who can use it: Available to API teams using Chat Completions and Responses API; unavailable for multi-agent models and some deferred requests.
How to use it: Add the `guidelines` array to the request body; for each item, send the `type: "policy"`, relevant `category`, and `behavior: "decline"` fields.
Tip: Send only the categories you want to change; unspecified categories retain their default behavior.
Gemini API CLI beta released
Google has released in beta the stateless, dependency-free gemini-api command-line client, which lets users make Gemini API calls from a shell, coding agent, or CI pipeline. The tool supports text and media generation, file analysis, and managed agent execution.
Usage details
Who can use it: Developers with a Gemini API key; installation options are available for macOS, Linux, and Windows.
How to use it: Install the gemini-api CLI and set the GEMINI_API_KEY environment variable; use the generate, analyze, image, or agent commands.
Tip: Pin a specific CLI version in automations; beta releases may introduce backward-incompatible changes.
✓ Verified by official sourceSource: Release notes | Gemini API - Google AI for DevelopersSource: Gemini API CLIiCloud Mail and Calendar plugins are now available
iCloud users can now connect Mail and Calendar to ChatGPT to find emails, summarize their inbox, check their schedule, and update calendar events. This extends ChatGPT’s personal productivity workflow to Apple services.
Usage details
Who can use it: All ChatGPT plans; on web, desktop app, and iOS.
How to use it: Open the apps or plugins section in ChatGPT and connect iCloud Mail and iCloud Calendar.
Tip: Before making calendar changes, verify the write permission requested by the plugin and the event details.
✓ Verified by official sourceSource: ChatGPT yayın notlarıMuse launches in Singapore
Meta has launched its personal agent, Muse, in Singapore. Users can direct shopping, dining, travel and everyday tasks through connectors for Gmail, Google Calendar and Instagram, as well as Shopee, Grab, Singapore Airlines, Agoda, ShopBack, Klook, Trip.com and Lazada.
Usage details
Who can use it: Available in Singapore on iOS, Android and muse.ai; basic use is free, with subscription plans available for higher usage.
How to use it: Open the Muse app or muse.ai and connect the relevant apps and connectors in settings.
Tip: Before making a purchase or reservation, ask Muse to compare prices, options and terms.
✓ Verified by official sourceModels: Muse SparkSource: Muse Is Now Available in Singapore: The World’s First Personal AI Agent Built for EveryoneQwen-Audio-3.1-Realtime-Flash released
Qwen-Audio-3.1-Realtime-Flash offers bidirectional, real-time conversation with text and audio input and output. Support for function calling, web search, and voice cloning, along with a 262.144-token context window, delivers a significant upgrade for voice agents.
Usage details
Who can use it: Users covered by Model Studio’s International scope; API usage is subject to charges.
How to use it: Select qwen-audio-3.1-realtime-flash as the model ID in Model Studio and connect to the real-time audio API.
Tip: Enable function calling for voice customer service and have the agent confirm the outcome of an action during the conversation.
✓ Verified by official sourceModels: qwen-audio-3.1-realtime-flashSource: Model lifecycle and updatesClaude for Financial Advisors connector and skill package
The new package for financial advisors adds connectors to custodians, portfolio platforms, CRMs, and planning tools. Ready-to-use Skills for meeting preparation, portfolio rebalancing, compliance checks, and follow-up notes are also included in a single installation.
Usage details
Who can use it: Financial advisor teams and organizations; connector access depends on the relevant provider and Claude organization permissions.
How to use it: Install the Claude for Financial Advisors plugin from Claude Marketplace and select the providers you’ll use during setup.
Tip: Limit recommendations from every Skill that works with portfolio or client data to the advisor’s final decision.
✓ Verified by official sourceSource: Claude for Financial Advisors duyurusuCreate and configure a Meta app with an agent
Social Technologies MCP has moved beyond providing API and integration information and can now handle key steps in the app lifecycle through an agent. The agent guides users through creating an app, finding or creating a Business Portfolio, linking it, and adding use cases such as WhatsApp Business messaging.
Usage details
Who can use it: Rolling out gradually; it may not be available to all developers at the same time.
How to use it: Reconnect the MCP connection and ask the agent you use to create a new Meta integration; approve write operations one by one.
Tip: Do not share your app secret in the agent chat; Meta requires you to retrieve it from App Dashboard.
✓ Verified by official sourceSource: Developers can now create and configure an app for Social Technologies Dev Center using their agentGPT-6.1 Terra launches in Work, Codex, and the API
GPT-6.1 Terra launched as a fast, efficient model for visual reasoning, precise computer use, and natural writing. In addition to ChatGPT Work and Codex, it is available in the Responses API and Chat Completions API under the `gpt-6.1-terra` identifier.
Usage details
Who can use it: Phased rollout in ChatGPT Work and Codex to Plus, Pro, Business, Enterprise, and Edu plans; available to developers in the API.
How to use it: Select GPT-6.1 Terra from the model list in ChatGPT Work or Codex; use `gpt-6.1-terra` in the API.
Tip: For visual and computer-use workflows, evaluate Standard and Fast modes separately to measure the latency-cost tradeoff.
✓ Verified by official sourceModels: GPT-6.1 Terra, gpt-6.1-terraSource: ChatGPT yayın notlarıSource: OpenAI API changelogSubscribers receive a 25% quota refund after K3.1 Pro launch issues
Moonshot announced a one-time 25% increase to the monthly quota for all subscribers in response to rate-limiting and unnecessary quota consumption issues during the Kimi K3.1 Pro launch. It also changed parameters to reduce unnecessary subagent calls and output length.
Usage details
Who can use it: Kimi subscribers; quota is reset for users whose usage is below 25%.
How to use it: No action is required; check your account quota on the Kimi subscription or usage screen.
Tip: When checking your quota, also track how changes to subagent calls after launch affect task consumption.
✓ Verified by official sourceModels: Kimi K3.1 ProSource: Moonshot resets subscriber quotas after Kimi K3.1 Pro launch issuesAPI requests are stored by default in new paid projects
In new projects on the paid tier, Generate Content API requests are now stored by default with store=true. The setting for existing projects is unchanged; users can opt out by sending store=false per request or turning off logging in AI Studio Logs and Datasets settings.
Usage details
Who can use it: New Gemini API projects on the paid tier; existing projects retain their previous settings.
How to use it: Add store=false to the request body or turn off Generate Content API logging in AI Studio > Logs and Datasets > Settings.
Tip: For production calls containing sensitive data, explicitly set the storage preference before deployment.
✓ Verified by official sourceSource: Release notes | Gemini API - Google AI for DevelopersClaude for Nonprofits plugin released
The new plugin for verified nonprofit organizations combines organization mapping, AI policy checks, donor and grant research, proposal writing, marketing, and hiring Skills with connected tools. The plugin is installed automatically at no extra cost on eligible Team and Enterprise plans.
Usage details
Who can use it: Verified nonprofit organizations; installed automatically on Team and Enterprise plans at no extra cost. A separate discount was announced for verified nonprofits in low- and middle-income countries.
How to use it: Ask Claude to map where your organization stores its information; other nonprofit Skills use this map.
Tip: Before submitting grant applications, check figures cited by Claude against the original documents.
✓ Verified by official sourceSource: Claude for Nonprofits genişleme duyurusuRouting screen added for corporate IP restrictions
Kimi Work 3.2.21 displays a screen explaining why access was blocked by a corporate IP allowlist, with options to return to the personal workspace or try again.
Usage details
Who can use it: Kimi Work accounts using an IP allowlist managed by corporate workspace administrators.
How to use it: On the access restriction screen, choose Try again or Return to personal workspace.
Tip: For teams working outside the corporate network, separate personal and corporate workspaces in advance.
✓ Verified by official sourceSource: Kimi Work Release NotesAgent Swarm no longer starts automatically unless the user enables it
With Kimi Code 2.3.2, Agent executes tasks directly by default; the subagent swarm starts only if the user explicitly requests it or multi-agent mode is enabled.
Usage details
Who can use it: Kimi Code CLI 2.3.2, Desktop 1.1.3, and relevant web sessions.
How to use it: Enable Swarm mode with the /swarm command or in the multi-agent mode settings.
Tip: Leave Swarm off for simple tasks to reduce quota consumption; enable it deliberately for parallel research.
Auto-review rules can be managed from cards
Grok Bot can now suggest adding, changing, or removing an Auto-review rule from the in-app approval card. Because Auto-review performs a second check, it also reduces unnecessary approval cards for actions deemed safe.
Usage details
Who can use it: For accounts using Grok Bot; the relevant desktop app version must be 0.75.2 or later.
How to use it: Open and approve the Auto-review rule suggestion shown on an approval card; rules can also be edited in Grok Bot settings.
Tip: Add rules for repeated, low-risk actions, but keep irreversible actions such as sending money or external messages subject to manual approval.
✓ Verified by official sourceSource: Grok Bot Changelog3D Make launches on the web and in the Kimi app
Alongside the Kimi K3.1 Pro announcement, 3D Make went live on kimi.com and in the Kimi app. Users can preview and edit 3D models online, download 3D-print-ready files, or export the result as a video or Live Photo.
Usage details
Who can use it: Users of Kimi.com and the Kimi mobile app; access may depend on account and regional support.
How to use it: Open 3D Make in Kimi, provide text, a sketch, or a reference image; then preview, edit, and export the model.
Tip: Verify dimensions and geometry in desktop 3D software before 3D printing.
K3 series API pricing, automatic caching, and Batch support announced
The API pricing table lists K3.1 Pro at $2 for input and $10 for output; Highspeed is priced at $4 and $20/M tokens. Context caching is automatic for the K3 series, with 5-minute or 1-hour TTL options; K3.1 Pro and K3 Batch API run at 60% of the standard price.
Usage details
Who can use it: Pay-as-you-go Kimi API customers; prices are per million tokens and exclude taxes.
How to use it: Choose a model; for requests with long intervals, use the 1h option with prompt_cache_options.ttl; switch to Batch API for batch jobs.
Tip: Increasing cache hits for repeated long system prompts can significantly reduce costs.
✓ Verified by official sourceModels: Kimi K3.1 Pro, Kimi K3.1 Pro Highspeed, Kimi K3Source: Model Inference PricingSource: Kimi K3.1 Pro QuickstartMandatory thinking and reasoning_effort setting added to K3.1 Pro API
K3.1 Pro API calls now always keep thinking mode enabled and allow users to select the low, high, or max level with the reasoning_effort field. The model also supports image input, tool calls, and a 1-million-token context window.
Usage details
Who can use it: Developers using Kimi API; turning off thinking is not supported for K3.1 Pro.
How to use it: Send the top-level reasoning_effort field in the request as low, high, or max; choose the kimi-k3.1-pro or highspeed model ID.
Tip: Start with low to reduce response latency, and move up to high or max for complex planning and coding tasks.
✓ Verified by official sourceModels: Kimi K3.1 Pro, Kimi K3.1 Pro HighspeedSource: Kimi K3.1 Pro QuickstartKimi K3.1 Pro launches as a new 3T-class flagship model
Moonshot released Kimi K3.1 Pro, a new open-weight model with a 1-million-token context window, developed using greater post-training compute than K3. Improvements are aimed at coding, 3D generation, financial research, and long-horizon Agent tasks.
Usage details
Who can use it: Kimi.com, Kimi mobile apps, Kimi Work, Kimi Code, and Kimi API; Kimi Enterprise administrators must enable access separately.
How to use it: Use /model in Kimi Code, the kimi-k3.1-pro ID in the API, or the model picker in the app and on the web.
Tip: For long tasks, set the thinking level to max and use the extended context only when truly necessary.
✓ Verified by official sourceModels: Kimi K3.1 ProSource: Introducing Kimi K3.1 ProSource: What's New | Kimi Code DocsKimi K3.1 Pro Highspeed offers a 2–3x speed option at the same quality
Kimi K3.1 Pro Highspeed became available, retaining the same model quality as K3.1 Pro while delivering output approximately 2–3x faster.
Usage details
Who can use it: Kimi.com, Kimi Work, Kimi Code, and Kimi API; Pro/Allegretto or higher access in Kimi Code.
How to use it: Use the kimi-k3.1-pro-highspeed model ID in the API or the model picker in Kimi Code.
Tip: Choose Highspeed for interactive code fixes and rapid prototyping, and standard K3.1 Pro for in-depth analysis.
✓ Verified by official sourceModels: Kimi K3.1 Pro HighspeedSource: Introducing Kimi K3.1 ProSource: Kimi K3.1 Pro QuickstartVault, email forwarding, and higher routine limits
Bots can now save login details, cards, and contact information to Vault in Settings and use them to fill out forms; they can forward emails from their own inboxes, preserving formatting, images, and attachments. The total routine run limit has increased from 100 to 300 per hour, and the per-routine limit is now 200.
Usage details
Who can use it: For Grok Bot users; Vault and bot email features are available in version 0.75.2 or later.
How to use it: Open Vault from Settings; ask a Bot to fill out a login or payment form, or forward an email from its own inbox.
Tip: Save only the information in Vault that automation actually needs, and keep Auto-review approval enabled for payment submissions.
✓ Verified by official sourceSource: Grok Bot ChangelogClaude Tag API beta opens
The beta Claude Tag API for Claude Enterprise organizations provides endpoints for reading and changing Claude Tag settings in Slack channels and workspaces. You can see the source of channel settings and attach Access bundles.
Usage details
Who can use it: Claude Enterprise organizations; requires an Admin API key, the read:claude_tag or write:claude_tag scope, and the experimental-claude-tag-2026-10-21 beta header. Not available for Claude Console organizations.
How to use it: Call the /v1/organizations/claude_tag/ endpoints with an appropriate Admin API key.
Tip: Before changing channel settings, use the API to check whether the active value comes from the workspace or the channel.
✓ Verified by official sourceSource: Claude Platform API sürüm notlarıAutoClaw unlocks a 40M-token weekly reward with a single login
During the October 26–November 1 UTC+8 event, users can receive a 40M-token reward worth 4,000 credits by logging into their account once. The reward expires on November 1 at 23:59 UTC+8.
Usage details
Who can use it: AutoClaw users can claim it once during the event; the reward is still active as of the October 31 check date.
How to use it: Log in to AutoClaw between October 26 and November 1 and claim the weekly reward; each account can claim it once, and the credits must be spent before they expire.
Tip: Claim the reward as early as possible; even if it overlaps with the next campaign, don't miss the November 1 deadline for the previous 40M credits.
✓ Verified by official sourceSource: Haftalık ayrıcalık: girişle 40M tokenAnthropic-managed Claude Models launches on Google Cloud
The Claude API now has a new deployment option operated by Anthropic and accessed through a Google Cloud project. Access is managed through Google Cloud IAM, while billing is handled through Google Cloud Marketplace; Fast mode for Claude Opus 5.5 is also available through this channel.
Usage details
Who can use it: Developers with Google Cloud projects and an eligible Anthropic-managed Claude Models setup.
How to use it: Enable Anthropic-managed Claude Models in your Google Cloud project, then configure IAM permissions and Marketplace billing.
Tip: Compare Anthropic API, Google-managed, and Anthropic-managed options based on your latency, IAM, and billing requirements.
Bots can create Pages and videos
Grok Bot can now create Pages, such as reports, notes, and dashboards, that can be edited collaboratively or kept up to date. It can also generate videos from a description or a shared image, taking bot output beyond chat text.
Usage details
Who can use it: For users with Grok Bot access; sharing permissions are subject to team administrator policies.
How to use it: In a Bot conversation, ask it to create a Page in report or dashboard format, or a video from a description or image; use the sharing menu for Pages.
Tip: For a report that needs ongoing updates, use a Page as a living workspace rather than a one-off file.
✓ Verified by official sourceSource: Grok Bot ChangelogDecisions API becomes generally available
Powered by GPT-6 Luna, the Decisions API converts text or images into typed responses from predefined options. The general release adds shared input caching, an image per question, `input_json`, and capacity for up to 400 questions per request; this workflow progressed from the late-September preview and October 6 beta stages.
Usage details
Who can use it: OpenAI API developers; became generally available on October 22.
How to use it: Use the `v1/decisions` endpoint and the `gpt-6-luna` model; provide `input_json` for structured context.
Tip: Define classification or routing options in advance as a limited, mutually exclusive schema.
✓ Verified by official sourceModels: GPT-6 Luna, gpt-6-lunaSource: Decisions API duyurusuSource: OpenAI API changelogMeet speech translation expands to 63 languages
Gemini-powered Google Meet speech translation now supports 63 languages on web and mobile, automatic detection of the spoken language, and independent language preferences for each participant. Audio and synchronized caption translation are available even when multiple languages are spoken in the same meeting.
Usage details
Who can use it: Business Standard/Plus, Enterprise Standard/Plus, Google AI Pro/Ultra, Frontline Plus, Google AI Pro for Education, AI Expanded Access and AI Ultra Access; regional and age restrictions apply.
How to use it: During a meeting, open the translation control in the Meet toolbar; administrators can manage the feature via Apps > Google Workspace > Google Meet > Gemini settings.
Tip: For multilingual meetings, ask each participant to select their preferred target language in advance.
✓ Verified by official sourceModels: Gemini 3.5 Live TranslateSource: Real-time speech translation in Google Meet expands to 63 languages with automatic language detectionQwenCloud and region selection added to the Qwen Code auth flow
QwenCloud sign-in is now visible on the initial /auth screen; pay-as-you-go and Token Plan keys can be selected separately, and the next step lets users choose the China or International region. This makes Qwen Code’s Qwen Cloud/Model Studio credentials easier to distinguish.
Usage details
Who can use it: Developers using QwenCloud; China and International accounts/keys are separate.
How to use it: Start the /auth flow in Qwen Code, then select the key type and China/International region.
Tip: Make sure the API key’s region matches the workspace endpoint.
Qwen Code can connect to custom WeCom apps
A company’s custom-built WeCom app can now connect as a Qwen Code Channel. Employees send a task to the app, Qwen Code carries it out, and the response is sent back under the same app identity.
Usage details
Who can use it: Teams that develop enterprise WeCom apps; relevant deployment support is required.
How to use it: Add your WeCom app to the Channel configuration and define the permissions needed to receive and reply to messages.
Tip: In production, limit the app identity’s scope by granting only the necessary group and user permissions.
Unified audio generation with Qwen-Audio-3.1-TTS-Next
Rather than synthesizing speech alone, Qwen-Audio-3.1-TTS-Next can combine speech, effects, ambient audio, and multi-speaker content in a single pass. It brings production workflows such as podcasts, short videos, cinematic sound design, and game audio together in one model.
Usage details
Who can use it: Available to Model Studio International users; API access and pricing depend on account/region configuration.
How to use it: Call the qwen-audio-3.1-tts-next model through the Model Studio API; include text, timestamps, and reference audio data in the request.
Tip: Combine dialogue and ambient audio in the same generation to create drafts that don’t require separate mixing afterward.
Zoho CRM and enhanced GitHub, Gmail, and Calendar connections
Grok Bot’s connector lineup has expanded: Zoho CRM now offers regional data center selection, and GitHub can now read repository, commit, pull request, and comment details. Gmail and Google Calendar connections also support richer actions, such as attachments, reply-all, and creating multiple events.
Usage details
Who can use it: For Grok Bot users; each connector’s own account, region, and permission requirements apply.
How to use it: Connect Zoho CRM, GitHub, Gmail, or Google Calendar from Connect Apps; ask the Bot to perform the relevant action in a chat.
Tip: Select the correct regional data center for your account before signing in to Zoho CRM; choosing the wrong region may cause connection issues.
✓ Verified by official sourceSource: Grok Bot ChangelogAgent Skills standard transferred to Agentic AI Foundation
Anthropic donated the name, repository, and domain of the Agent Skills standard—which gives agents specialized knowledge and workflows—to the Agentic AI Foundation. Making the standard vendor-neutral and community-maintained supports Skills working with more than 70 agent platforms beyond Claude.
Usage details
Who can use it: The Agent Skills standard is publicly available and can be used with Claude and other agent platforms that support the standard.
How to use it: Create Skills based on SKILL.md and share them according to the agentskills.io standard.
Tip: For enterprise Skills, keep instructions, scripts, and reference files separate and versioned.
✓ Verified by official sourceSource: Agent Skills’in Agentic AI Foundation’a bağışıClaude-format plugin commands run in the Qwen Code terminal
Commands packaged in Claude-format plugins can now run in the terminal without requesting a model call. When the Mods switch is on, a plugin can modify an argument or block a call before a tool call; the feature is off by default.
Usage details
Who can use it: Qwen Code v0.26.0; the experimental Mods setting and workspace trust are required.
How to use it: Turn on Mods in Settings and restart; trust the workspace before using a plugin.
Tip: Before enabling a third-party plugin, review which files and tools its commands can access.
MiMo Studio expands format and device compatibility for audio input
MiMo Studio V3.4.2 now supports more audio input formats and parsing compatibility across more browsers and devices. The changelog does not list the names of the newly supported formats.
Usage details
Who can use it: Users who use audio input in MiMo Studio; the official changelog does not mention any plan- or region-based restrictions.
How to use it: Open audio input or the related MiMo Chat/VoiceClone experience in MiMo Studio, then upload or record an audio file. Since no specific format list is provided, use a file type accepted by the interface.
Tip: If the file cannot be parsed, try again in an up-to-date browser, or convert the same recording to another common audio format and upload it again.
✓ Verified by official sourceSource: Xiaomi MiMo Studio özellik günlüğü (V3.4.2)Qwen Code creates editable interactive cards
When Generated UI is enabled, Qwen Code creates cards whose numerical values can be edited directly and whose totals are recalculated instantly. This makes it possible to experiment with budgets, planning, and capacity comparisons without sending a new message for every change.
Usage details
Who can use it: Qwen Code v0.25.3/v0.26.0; disabled as an experimental setting and not yet available in managed/remote/sandbox sessions.
How to use it: Turn on Generated UI in Experimental settings and restart Qwen Code.
Tip: Adjust variables such as rent, per-person costs, or server traffic in the card to compare alternatives side by side.
Qwen Code finds workflows in the root directory from subdirectories
Test and validation workflows saved at the Git repository root can be found by searching upward and used even when a session starts in a subdirectory. This reduces the need to redefine the same project checks in frontend, backend, and component directories.
Usage details
Who can use it: Local terminal and standard Web Shell sessions; directories within the same trusted Git repository.
How to use it: Open the project from a subdirectory; workflows saved at the root will automatically appear in the session list.
Tip: Keep a single validation workflow at the root instead of copying a separate workflow for each component.
Qwen Code QQ bot can receive and read files
The QQ bot can now save and read Word, PDF, and spreadsheet files sent directly in a message or with an @ mention in a group. It bases its response on the file’s contents, expanding document-processing workflows in the chat channel.
Usage details
Who can use it: Qwen Code Channel users with a QQ bot installed.
How to use it: Send a supported file directly to the QQ bot or mention the bot in a group.
Tip: Control token usage by uploading a long document and first asking it to extract sections or fields, then to summarize.
Manage Web Shell services with one command group using qwen daemon
The qwen daemon start, status, logs, stop, and restart commands bring local Web Shell services together under a single management interface. --detail shows the version, uptime, and session count; --wait prevents the next automation step from running until the service is ready.
Usage details
Who can use it: Local Qwen Code Web Shell users.
How to use it: Start with qwen daemon start; manage the service with qwen daemon status, logs, or restart --port <port>.
Tip: Use start --wait in CI or local scripts to avoid starting tests before the service is ready.
SpaceX Super Intelligence Partner Network launches
xAI has launched an official Partner Network program for companies building products, solutions, and services with Grok. The program brings together enterprise transformation partners and Marketplace partners developing add-ons and product integrations for the Grok product family.
Usage details
Who can use it: Open to applications from enterprise customers and companies building solutions around Grok; partnership access is subject to application and evaluation.
How to use it: Review and apply to the Transformation Partners or Marketplace Partners program at x.ai/partners.
Tip: If you’re planning an enterprise rollout, clarify upfront whether you fall under product integration or consulting/implementation services.
✓ Verified by official sourceSource: Introducing the SpaceX Super Intelligence Partner NetworkVault adds filtered exports for Gemini conversations
Google Vault can now filter Gemini in Workspace conversations in Gmail, Drive, Docs, Sheets, Slides, and Meet by user/OU and date range, and export them as JSON. Exports can include prompts, responses, attachments, links, and citations.
Usage details
Who can use it: Organizations with Business Plus, Enterprise Standard/Plus, Frontline Standard/Plus, Enterprise Essentials Plus, Education Fundamentals/Standard/Plus, and the Vault add-on.
How to use it: In Vault > Search, select Gemini in Workspace from the service list, then start an Export using custodian user/OU and date filters.
Tip: Limit the export scope to the smallest necessary OU and date range to avoid collecting unnecessary sensitive data.
✓ Verified by official sourceSource: Google Vault now supports filtered exports for Gemini in WorkspaceNotice added on the purpose of audio data use for VoiceClone uploads
MiMo Studio V3.4.2 added a new notice in the VoiceClone experience informing users what uploaded voice data will be used for. The change provides clearer information before audio data is uploaded.
Usage details
Who can use it: Users of the MiMo Studio VoiceClone experience.
How to use it: Open VoiceClone in MiMo Studio; before uploading an audio sample, read the notice about the purpose of using voice data as sensitive information and complete the confirmation flow.
Tip: Upload only necessary audio samples you are authorized to use; obtain explicit consent before using someone else’s voice.
✓ Verified by official sourceModels: mimo-v2.5-tts-voicecloneSource: Xiaomi MiMo Studio özellik günlüğü (V3.4.2)API version 2026-10-21 and new SDK/CLI series released
Anthropic released API version 2026-10-21 and made the new major versions of its Python, TypeScript, C#, Go, Java, PHP, and Ruby SDKs, along with ant CLI, default to this version. The new versions include breaking changes independent of the API, and the license changed from MIT to Apache 2.0.
Usage details
Who can use it: Claude Platform developers; the API version is available for Claude 4.7 and later.
How to use it: Apply the SDK and client changes in the Migration to API version 2026-10-21 guide.
Tip: Before upgrading the client in production, compare both the old and new API versions using the same test suite.
✓ Verified by official sourceSource: Claude Platform API sürüm notlarıSource: 2026-10-21 API sürümüne geçişClaude Duet proactive agent research preview
Claude Duet launched as a separate app that checks connected apps every few hours, messages users when something important comes up, and continues requested tasks while they’re away. It asks for approval before impactful actions such as sending messages, making payments, signing documents, or deleting items.
Usage details
Who can use it: Research preview and waitlist; initially for Pro and Max subscribers in the US, on iOS, Android, and claude.ai.
How to use it: Join the Claude Duet waitlist and connect your apps with an approval flow.
Tip: Start with reversible, low-risk tasks and clearly define which situations should trigger a notification.
Sunset date announced for marketing messages via Cloud API
Meta announced that it will discontinue marketing messages sent through WhatsApp Cloud API on October 13, 2027. Businesses must move their marketing traffic to the new Marketing Messages API for WhatsApp endpoint; utility messages that are reclassified will also be routed through it.
Usage details
Who can use it: Existing Cloud API integrations will continue working until October 13, 2027; the transition affects all WhatsApp business developers.
How to use it: Accept the Marketing Messages API for WhatsApp terms, validate the live integration on the new endpoint, and update routing for reclassified traffic.
Tip: Do not wait until the deadline to transition; Meta says most integrations take about two weeks.
Docs Live creates structured documents by voice on mobile
Docs Live is beginning to roll out to eligible Workspace users in the Google Docs apps for Android and iOS. Users can describe their ideas in natural speech; Gemini uses context from Gmail, Drive, and Chat to draft a structured document.
Usage details
Who can use it: Business Standard/Plus, Enterprise Standard/Plus, Google AI Pro/Ultra, Google AI Pro for Education and AI Expanded Access.
How to use it: Open Docs Live in the Google Docs mobile app and describe your document’s purpose and the context to use aloud.
Tip: In your voice prompt, specify the target document type, audience, and delivery format from the outset.
✓ Verified by official sourceModels: Gemini 3.8 LiveSource: Create structured documents using just your voice with Docs Live in Google Docs on mobileEnterprise users can create custom plugins and upload them to the marketplace
Kimi Work 3.2.18 enabled Enterprise users to create custom plugins and upload them to the plugin marketplace.
Usage details
Who can use it: Kimi Work Enterprise workspaces.
How to use it: In an Enterprise workspace, start the custom plugin creation or upload flow from the plugin marketplace.
Tip: Test organization-specific MCP connections in a limited workspace first, then publish them to the marketplace.
✓ Verified by official sourceSource: Kimi Work Release NotesCentral Customize hub launches in Gemini
The new Customize hub in the Gemini app lets users manage personal skills and more than 40 connected apps in one place. Users can adjust response tone, format, and connected services from the same menu.
Usage details
Who can use it: Rolling out globally to all user tiers on web, mobile, and Windows/macOS desktop apps; many connected apps are English- and US-focused, while some are available in additional English-speaking regions, Brazil, and India.
How to use it: Open the left navigation menu in Gemini and go to Customize; adjust settings in the Skills or Connected apps sections.
Tip: Regularly remove app connections you don’t use and keep only the necessary permissions enabled.
✓ Verified by official sourceSource: An easier way to customize how Gemini works for youMiMo V2.5-Pro and V2.5 model IDs retired
The model names `mimo-v2.5-pro` and `mimo-v2.5` became invalid as of October 21, 2026 10:00 (UTC+8). Requests using these names now return an error; Xiaomi is asking integrations to migrate to the V2.6 series.
Usage details
Who can use it: API integrations using the old V2.5 model IDs no longer work; the new V2.6 model IDs are available in the API.
How to use it: In code and configuration, replace `mimo-v2.5-pro` with `mimo-v2.6-pro` and `mimo-v2.5` with `mimo-v2.6-flash`. Verify the current IDs in the API model list and retest failed requests.
Tip: Manage model names through a single configuration variable instead of hard-coding them in application code; this lets you migrate from one place when future retirements occur.
✓ Verified by official sourceModels: mimo-v2.5-pro, mimo-v2.5, mimo-v2.6-pro, mimo-v2.6-flashSource: MiMo model emeklilik günlüğüSource: MiMo model listesi (19 Ekim güncellemesi)Source: Sina Tech / IT之家: MiMo-V2.5-Pro bugün kullanımdan kaldırıldıQwen-Audio-3.1-TTS-Plus launched
Qwen-Audio-3.1-TTS-Plus adds support for multiple languages and dialects, streaming synthesis, free-form instructions, and tags for emotion, tone, persona, speed, and volume. It focuses on professional dubbing and brand voice use, with improved resistance to noise and echo in voice cloning.
Usage details
Who can use it: For API users on Model Studio International.
How to use it: Use the qwen-audio-3.1-tts-plus model and audio control tags in the request body.
Tip: Generate the same text with different persona and emotion tags to create a short comparison set for your brand voice.
ChatGPT Ads expands to 26 more countries
ChatGPT Ads has expanded to 26 new countries, bringing its total availability to 89 countries. The expansion covers new markets in Latin America, as well as Europe, the Middle East, Africa and Asia.
Usage details
Who can use it: The market expansion was announced for advertisers; exact eligibility varies by Ads Manager account and country.
How to use it: Apply for ChatGPT Ads through Ads Manager using an advertiser account in an eligible country.
Tip: Before launching a campaign, check that the target country is enabled in Ads Manager and review local advertising policies.
✓ Verified by official sourceSource: ChatGPT Ads 26 yeni ülkeye genişlediSwitching between Enterprise workspaces is now easier
In Work 3.2.18, users can switch between workspaces grouped by Enterprise from the avatar menu, and user role labels are displayed.
Usage details
Who can use it: Kimi Work users connected to multiple Enterprise workspaces.
How to use it: Open the avatar menu and select the target workspace from the Enterprise group.
Tip: Check the active workspace label before starting a task to avoid accidentally sending personal files to a corporate workspace.
✓ Verified by official sourceSource: Kimi Work Release NotesA review step comes to Notebook presentation and video generation
Gemini Notebook has added a step-by-step flow for guiding choices about focus, format, and visual style before generating a Slide Deck or Video Overview. When Review outline or Review producer notes is enabled, users can review the final draft and provide feedback.
Usage details
Who can use it: Google Workspace, Workspace Individual and personal Google account users.
How to use it: In the Notebook Studio panel, select Slide Deck or Video; enable Review outline or Review producer notes.
Tip: For long reports, review the first draft’s structure and narrative tone before generating to reduce the cost of regenerating.
✓ Verified by official sourceSource: Gain more creative control over Slide Decks and Video Overviews in Gemini NotebookPlugin skills now use qualified names
Kimi Code CLI 2.3.0 switched to calling plugin skills in the pluginID:skillName format to prevent skills with the same name from shadowing one another.
Usage details
Who can use it: Kimi Code CLI 2.3.0 users and plugin developers.
How to use it: Update skill calls to the <pluginID>:<skillName> format.
Tip: Scan existing AGENTS.md files and automations, and update old unqualified skill names before upgrading.
✓ Verified by official sourceSource: Kimi Code CLI ChangelogAPI service tiers can be restricted for projects
A new setting lets users choose which service tiers can be used on a per-project basis. Organization settings can restrict the service tiers allowed across all projects, enabling centralized enforcement of cost and latency policies.
Usage details
Who can use it: OpenAI API organization and project administrators.
How to use it: Configure service tier permissions at the project or organization level in Platform settings.
Tip: In production projects, disable high-cost tiers such as Ultrafast by default and enable them only for projects that need them.
✓ Verified by official sourceSource: OpenAI API changelogRemote Control streaming and file transfer limits expanded
CLI 2.3.0 added streaming support for Remote Control sessions and raised the transfer limit from 10 MiB to 100 MiB.
Usage details
Who can use it: CLI users using Kimi Code Remote Control.
How to use it: Start a Remote Control connection and watch the live stream from the connected device; use the 100 MiB limit for larger transfers.
Tip: On a mobile connection, test with text and small files first to monitor data usage.
✓ Verified by official sourceSource: Kimi Code CLI ChangelogUser profile enforcement fields and filter added
The enforcement_status and enforcement_actions fields have been added to User profile records. Developers operating platforms or gateways can use the enforcement_status[] filter to identify which end customer has been restricted or banned.
Usage details
Who can use it: Users of the User Profiles API on Claude Platform.
How to use it: Use the enforcement_status[] filter on the GET /v1/user_profiles endpoint.
Tip: Before applying restrictions across an organization, log profile records associated with end-customer IDs.
✓ Verified by official sourceSource: Claude Platform API sürüm notlarıSubscription payments with MoMo begin in Vietnam
Users in Vietnam can pay for ChatGPT Go, Plus, and Pro subscriptions through chatgpt.com using MoMo. Mobile app subscriptions continue to be managed by the Apple App Store or Google Play.
Usage details
Who can use it: Vietnam; only for Go, Plus, and Pro subscriptions purchased through chatgpt.com.
How to use it: When purchasing a subscription on the web, select MoMo from the payment methods.
Tip: For subscriptions purchased on iOS or Android, manage the payment method through your app store account.
✓ Verified by official sourceSource: ChatGPT yayın notlarıTask process visibility is now configurable
Work 3.2.18 displays task execution time at the top by default and lets users choose whether to show the task process.
Usage details
Who can use it: Kimi Work 3.2.18 users.
How to use it: Change the task process view under Settings > Other > General.
Tip: Hide the process for tasks where you want to see the result quickly; keep it visible when debugging.
✓ Verified by official sourceSource: Kimi Work Release Notes12-hour prompt caching option for the API
The Responses API now offers a 12-hour prompt caching option to keep shared instructions, tools and context cached for hours. While this can improve cache hits for long-running agents, 12-hour writes are billed at 2.5 times the standard input token price.
Usage details
Who can use it: API customers using supported models in the Responses API.
How to use it: Select a 12-hour cache duration in the prompt caching configuration.
Tip: Cache long, frequently repeated system instructions and avoid unnecessary 12-hour write costs for short tasks.
✓ Verified by official sourceSource: OpenAI API changelogIncident management with IncidentWatch and IncidentSweep
Two teammates have been added to Grok Bot: IncidentWatch, which monitors increases in similar tickets in the support queue, and IncidentSweep, which applies approved bulk responses. With human approval preserved, the process is automated from incident detection through resolution of affected tickets.
Usage details
Who can use it: Announced as a sample workflow for Grok Bot users; production use depends on the relevant support system and permission connections.
How to use it: Connect the support queue and status page, then define periodic monitoring for IncidentWatch and scope, exclusions, and approval steps for IncidentSweep.
Tip: Have a person approve sample tickets, excluded records, and the exact response text before IncidentSweep sends anything.
✓ Verified by official sourceSource: Using Grok Bot to detect and respond to incidentsLocked chats are now available
Locked chats hide selected conversations from regular history, pinned chats and search. They can be opened with a 4-digit PIN or supported device authentication, without deleting them. Data retention and model improvement settings remain unchanged.
Usage details
Who can use it: Rolling out to Free, Go, Plus and Pro users on web, iOS, Android, and the new macOS/Windows desktop app.
How to use it: Open a chat’s three-dot menu and select Lock chat.
Tip: Remember that locking a chat does not exempt it from the data retention policy; review data controls separately for sensitive content.
✓ Verified by official sourceSource: ChatGPT yayın notlarıWeb fetch and web search domain controls added to organization settings
Organization administrators can turn off web fetch or specify allowed and blocked domains from the Capabilities page in Claude Console. In Managed Agents sessions, web_fetch and web_search now enforce both organization restrictions and the tool-level allowed_domains or blocked_domains list.
Usage details
Who can use it: Claude Console organization administrators and Managed Agents; if web fetch is disabled, an API request returns 400, while an agent call returns an error result.
How to use it: Configure Allow web search and web fetch, along with domain lists, under Claude Console > Capabilities.
Tip: Limit the allowlist to the hosts your agent actually needs instead of using broad wildcards.
✓ Verified by official sourceSource: Claude Platform API sürüm notlarıDeepSeek Recipe 0.2.0 released
DeepSeek Recipe 0.2.0 updated its Rust libraries and Python bindings to convert V4 and V4.1 requests based on the model version. The new release handles image, tool search, structured output, and reasoning options more accurately across Responses API, Chat Completions, and Messages streams.
Usage details
Who can use it: Available to developers as an open-source Rust/Python package.
How to use it: Upgrade to deepseek-recipe 0.2.0; use ConversionOptions.model_versions_by_name or with_model_alias for custom model names.
Tip: When migrating from 0.1, check the changed APIs for ConversationRequest and stream generators.
✓ Verified by official sourceModels: DeepSeek-V4-Pro, DeepSeek-V4.1-FlashSource: DeepSeek Recipe değişiklik günlüğüInference hooks exit beta for Enterprise
Inference hooks have moved out of research or beta preview and are generally available to Claude Enterprise organizations. Organizations can run observability and control logic around model calls through a more stable interface.
Usage details
Who can use it: Claude Enterprise organizations.
How to use it: Follow the Inference hooks setup in the Claude Platform documentation and configure hooks for your organization.
Tip: Measure latency and errors in hooks; test rules that could block critical production calls in shadow mode first.
✓ Verified by official sourceSource: Claude Platform API sürüm notlarıOrganization webhooks added in beta
Organization webhooks have been added to the Admin API to notify you when Anthropic restricts or bans a user profile, and when the restriction ends. This lets platform providers trigger compliance workflows based on events.
Usage details
Who can use it: Claude Platform organizations; contact the Anthropic account team and use the webhooks-2026-10-07 header for beta access.
How to use it: Register a webhook destination at the /v1/organizations/webhooks endpoints.
Tip: Implement signature verification, retries, and an idempotency key in your webhook receiver.
✓ Verified by official sourceSource: Claude Platform API sürüm notlarıGuided setup for Auto mode and Code Review levels
The /auto-mode-setup command has been added to Claude Code; it reads the project and recent sessions to prepare the automatic permission classifier for the environment. In the same release, Code Review administrators can choose Quick, Standard, and Deep levels at the organization or repository level.
Usage details
Who can use it: Claude Code 2.1.304 and later; Code Review levels are for Team and Enterprise organizations.
How to use it: Run /auto-mode-setup in a Claude Code session; in Code Review settings, select Quick, Standard, or Deep for the repository.
Tip: Run Auto mode setup on a trusted sample project, and don’t generalize its recommendations without reviewing the suggested rules.
✓ Verified by official sourceSource: Claude Code 2.1.304 sürüm notlarıGemini analyzes BigQuery data in natural language through Sheets
Users who connect to BigQuery with Connected Sheets can now query the entire BigQuery dataset in natural language, not just visible cells. Gemini can also create dashboards with refreshable formulas, pivot tables, and charts based on an approved plan.
Usage details
Who can use it: Business Standard/Plus, Enterprise Standard/Plus, Google AI Pro/Ultra, Google AI Pro for Education, and AI Expanded Access; BigQuery query permission is required.
How to use it: In a spreadsheet connected to BigQuery, open the Gemini side panel and ask a question or request a dashboard in natural language.
Tip: For large datasets, first verify the results with a sample query and date range, then generate the dashboard.
✓ Verified by official sourceSource: Analyze BigQuery data using natural language with Gemini in Google SheetsNew China subscription plans and shared credit pool clarified
The official plan view captured on October 19, 2026, listed Go, Plus, Pro, Max, and Ultra plans for China; Plus and higher include Kimi Code access, a shared monthly credit pool, and a rolling 5-hour usage limit with no weekly cap on the new plans.
Usage details
Who can use it: China subscription plans; Go does not include Kimi Code, while Plus and higher do. Existing Andante and higher plans retain the old rules.
How to use it: Check your plan and credit usage in Kimi via avatar → Settings → Subscription or in the app via Me → Membership Plan → Subscription.
Tip: Agent, Kimi Work, and Kimi Code draw from the same credit pool, so avoid piling long tasks into a single plan period.
✓ Verified by official sourceModels: Kimi K3.1 ProSource: Membership Pricing and Plan OverviewSource: Membership Credit Updates and Usage RulesClassroom Notebook gives teachers Gemini insight summaries
Teachers can view an overall engagement summary, common themes, and individual conversation transcripts from Gemini Notebook conversations assigned to students. On the same day, more detailed controls were also introduced for administrators to restrict external sharing of Gemini Notebook to trusted domains.
Usage details
Who can use it: Google Workspace for Education Fundamentals, Standard and Plus; only for Gemini Notebooks assigned through Classroom.
How to use it: The administrator must enable Gemini Notebook for the relevant OU/group; teachers should use the assigned Notebook’s insight summaries section in Classroom.
Tip: Before incorporating AI summaries into lesson plans, compare them against relevant student transcripts using a sample.
Gemini in Sheets comes to iOS
Gemini in the Google Sheets iOS app can now answer questions about data, generate analytical insights, and create charts. This brings the Gemini in Sheets experience on web and Android to iPhone and iPad.
Usage details
Who can use it: Business Starter/Standard/Plus, Enterprise Starter/Standard/Plus, Google AI Pro for Education and Google AI Plus/Pro/Ultra users.
How to use it: Open a compatible spreadsheet on iOS and tap the Gemini spark icon; enter a suggested prompt or your own.
Tip: First check the data structure with ‘Summarize this table’, then enter more targeted chart prompts.
✓ Verified by official sourceSource: Gemini in Google Sheets is now available on iOS devicesGemini Notebooks now available to Workspace users in the EEA
Google updated its Gemini Notebooks announcement on October 19, saying the feature is now available to Workspace users in the European Economic Area. The EEA was excluded from the previous version.
Usage details
Who can use it: All Google Workspace customers, including those in the EEA, Workspace Individual and personal Google account users; the administrator must enable access to Gemini and Gemini Notebook.
How to use it: Open New notebook from the left panel of the Gemini app; Workspace administrators should enable the Gemini and Gemini Notebook settings.
Tip: After the rollout in the EEA, also verify data sharing and the administrator’s OU/group policies.
✓ Verified by official sourceSource: Notebooks in Gemini: a dedicated workspace for focused, organized work, now for schools and organizationsInteractive apps in Gemini can be exported to Google Sites
Interactive apps created in the Gemini app can be exported with one click to a new, private Google Sites draft. The draft can be edited and published without copying code.
Usage details
Who can use it: Google Workspace, Workspace Individual and personal Google account users with access to the Gemini app; web, Android and iOS.
How to use it: Open the interactive app in Gemini and select Export > Google Sites.
Tip: Before publishing, check the sharing permissions and content visibility of the Sites draft.
✓ Verified by official sourceSource: Publish interactive apps from the Gemini app to Google SitesAutoClaw offers an 85M-token reward for daily logins
During the October 19–25 event, users could claim 3,000 credits (30M tokens) per day on October 19–20 and 500 credits (5M tokens) per day on October 21–25. Daily rewards expired at midnight UTC+8.
Usage details
Who can use it: It was open to all AutoClaw users; the event ran from October 19–25 UTC+8 and had ended as of October 31.
How to use it: Users had to log in to AutoClaw on event days and claim that day's reward. Each daily reward became invalid at 24:00 UTC+8 that day; the campaign is now closed.
Tip: For short-lived rewards, plan to claim and use them on the same day; don't leave them until the next day.
✓ Verified by official sourceSource: 85M token: günlük giriş etkinliğiThe 1017 release focused on end-to-end project generation
DeepSeek-V4.1-Flash-1017 strengthened end-to-end project generation for tasks such as game development and 3D modeling through self-feedback reinforcement learning. DeepSeek showcased a web-based racing game, a Portal 2 remake built with Godot, and a 3D model of the Forbidden City.
Usage details
Who can use it: DeepSeek web/app and compatible agent environments using the model.
How to use it: Specify the complete project goal, tools to use, and delivery criteria in a single prompt, and run it in an agent environment with file access.
Tip: Clearly define the working directory, dependency rules, and test commands before generating code and assets.
✓ Verified by official sourceModels: DeepSeek-V4.1-Flash-1017Source: DeepSeek-V4.1-Flash-1017 duyurusuSource: DeepSeek güncellemesini aktaran Silicon Republic haberiDeepSeek-V4.1-Flash-1017 open weights released
DeepSeek released the V4.1-Flash-1017 weights on Hugging Face and ModelScope. The Hugging Face model card lists the release as MIT-licensed and usable with image-to-text and text generation workflows.
Usage details
Who can use it: Developers with access to Hugging Face or ModelScope; local execution requires suitable hardware and inference infrastructure.
How to use it: Load the deepseek-ai/DeepSeek-V4.1-Flash-1017 model from Hugging Face with Transformers, vLLM, or SGLang.
Tip: Factor the model card’s approximately 763B parameters and 1M context requirements into your hardware plan.
✓ Verified by official sourceModels: DeepSeek-V4.1-Flash-1017Source: DeepSeek-V4.1-Flash-1017 duyurusuSource: DeepSeek-V4.1-Flash-1017 model kartıDeepSeek-V4.1-Flash-1017 strengthened for agentic tasks and coding
The 1017 update aims to improve reasoning, coding, agentic tasks, and visual benchmarks over the previous V4.1-Flash, while producing shorter outputs and reducing task duration. The DeepSWE v1.1 score rose from 74.2 to 77.0, while the cost-performance balance improved.
Usage details
Who can use it: Available via API through deepseek-flash and in the official web/app models.
How to use it: Run deepseek-flash calls without changing the model name; use an up-to-date agent environment for tool-based coding.
Tip: For long agent tasks, measure output length and total task duration to compare costs with the previous version.
✓ Verified by official sourceModels: DeepSeek-V4.1-Flash-1017, DeepSeek-V4.1-FlashSource: DeepSeek-V4.1-Flash-1017 duyurusuSource: DeepSeek-V4.1-Flash-1017 model kartıSource: DeepSeek güncellemesini aktaran Silicon Republic haberiDeepSeek-V4.1-Flash-1017 enabled on the web, in the app, and via API
DeepSeek upgraded V4.1-Flash to DeepSeek-V4.1-Flash-1017 and released the new version on its official website, in its app, and via API. The API call method and pricing have not changed; existing deepseek-flash integrations automatically use the new version.
Usage details
Who can use it: DeepSeek web and app users; accounts using deepseek-flash via API.
How to use it: Keep the model name as deepseek-flash in the API; no additional settings are needed on the app or web.
Tip: To track behavioral changes in results, compare important prompts again with outputs from before 1017.
✓ Verified by official sourceModels: DeepSeek-V4.1-Flash-1017Source: DeepSeek Değişiklik GünlüğüSource: DeepSeek-V4.1-Flash-1017 duyurusuSource: Models & PricingFetch API opened to everyone
DeepSeek has launched the Fetch API, which converts a single web page into Markdown content. This lets agents and RAG applications process a search result with a separate content retrieval call.
Usage details
Who can use it: Available to developers with a DeepSeek API key and sufficient balance.
How to use it: Call POST https://api.deepseek.com/v1/tools/fetch with a Bearer key and provide an HTTP/HTTPS url field in the JSON body.
Tip: Set up a small test workflow first, keeping in mind that failed calls or calls that produce no content are not charged.
✓ Verified by official sourceSource: DeepSeek-V4.1-Flash-1017 duyurusuSource: Search ve Fetch API rehberiSearch API opened to everyone
DeepSeek has opened its Search API, which uses the company’s in-house search engine, to all developers. The API returns structured web results with options for queries, result count, localization, and including page content.
Usage details
Who can use it: Available to developers with a DeepSeek API key and sufficient balance.
How to use it: Call POST https://api.deepseek.com/v1/tools/search with a Bearer key; query is required, count can be 1–50, and need_content is optional.
Tip: Use the site:, lang:, after:, and before: query operators to improve source quality.
✓ Verified by official sourceSource: DeepSeek-V4.1-Flash-1017 duyurusuSource: Search ve Fetch API rehberiPricing announced for Search and Fetch calls
Search and Fetch are billed per successful request rather than per token, at $0.75 per 1,000 successful calls for each endpoint. Billing for server-side calls in DeepSeek Harness, Claude Code, and Codex will begin on October 31, 2026 at 16:00 UTC.
Usage details
Who can use it: All developer accounts with an API balance; calls within tools are free until October 31, 2026 at 16:00 UTC.
How to use it: Call the separate Search or Fetch endpoint; requests that return a 200 status code and actual results are deducted from the balance.
Tip: Increasing the Search count does not change the price; cache results yourself to avoid unnecessary repeat calls.
✓ Verified by official sourceSource: DeepSeek-V4.1-Flash-1017 duyurusuSource: Models & PricingSource: Search ve Fetch API rehberiThe 1017 release was optimized for agent and coding tools
DeepSeek optimized the 1017 release for environments such as DeepSeek Harness, Tencent CodeBuddy, and WorkBuddy. DeepSeek Harness 0.2.1-rc.1 or later is recommended for the best experience.
Usage details
Who can use it: Developers using these agent and coding tools; the tools’ own access requirements apply.
How to use it: Upgrade DeepSeek Harness to 0.2.1-rc.1 or later and select the deepseek-flash model; use the DeepSeek API connection in CodeBuddy/WorkBuddy.
Tip: When comparing different agent environments on the same repository, keep tool permissions and the context window fixed.
✓ Verified by official sourceModels: DeepSeek-V4.1-Flash-1017Source: DeepSeek-V4.1-Flash-1017 duyurusuSource: DeepSeek Değişiklik GünlüğüSearch and Fetch calls in agent tools temporarily made free
Server-side Search and Fetch calls made through DeepSeek Harness, Claude Code, and Codex are free until October 31, 2026, at 16:00 UTC. After that, the same calls will switch to the standard per-request pricing.
Usage details
Who can use it: Users of DeepSeek Harness, Claude Code, or Codex who use the DeepSeek API; the free period lasts until October 31, 2026, at 16:00 UTC.
How to use it: Configure a supported agent tool with a DeepSeek API key and enable server-side web search.
Tip: Measure production traffic before November 1 and budget for Search and Fetch costs under the new pricing.
✓ Verified by official sourceSource: Search ve Fetch API rehberiSource: DeepSeek-V4.1-Flash-1017 duyurusuAccount-level QPS limits announced for Search and Fetch
The new Search API is limited to 100 requests per second per account, and the Fetch API to 25 requests per second. Fetch is also subject to an additional per-domain QPS limit and operates separately from model concurrency limits.
Usage details
Who can use it: Applies to all accounts using the Search and Fetch APIs.
How to use it: Queue requests below 100 QPS for Search and 25 QPS for Fetch; wait and retry on HTTP 429 responses.
Tip: Remember that multi-agent traffic shares the account-level quota; implement client-side backoff and retries.
✓ Verified by official sourceSource: Search ve Fetch API rehberiArtifact SDK and Ultrafast added to Agents API sandboxes
OpenAI-hosted Agents API sandboxes now include the Artifact SDK and skills for creating and editing spreadsheets, presentations and Word documents. The Agents API also adds a speed tier that reduces token latency on supported models using `agent.service_tier: ultrafast`.
Usage details
Who can use it: Developers using the Agents API; Ultrafast is available only on supported models and within eligible rate limits.
How to use it: Select a hosted sandbox environment; to use the speed tier, set `service_tier` to `ultrafast` in the agent configuration.
Tip: For document generation, edit the output with the Artifact SDK, then validate the formatting before delivering the final file to the user.
✓ Verified by official sourceSource: OpenAI API changelogCreate presentations collaboratively in ChatGPT
ChatGPT Slides lets users create presentations from a conversation, file or draft, and work on the same deck in real time with team members. Users can edit text, images, shapes and charts; present in ChatGPT; and export to PowerPoint, PDF or Google Slides.
Usage details
Who can use it: Rolling out on web and desktop to eligible Pro, Business and Enterprise users with Space access.
How to use it: In Work, ask ChatGPT to create a presentation in Space.
Tip: Add a sample deck or brand guidelines to the initial prompt to establish the layout and visual style from the outset.
GPT-6 Pro can run tasks in parallel with multiple agents
The GPT-6 Pro update in ChatGPT brings faster responses, better use of tools and plugins, and the ability to assign different parts of the same task to multiple agents working in parallel. Users can track progress in the conversation and open completed subtasks.
Usage details
Who can use it: For eligible ChatGPT Pro users; in managed workspaces, availability depends on administrator and deployment settings.
How to use it: Select GPT-6 Pro in the Chat tab and request a task that can be divided into parallel subtasks.
Tip: Break the task into clear, independent subtasks to speed up parallel work.
Custom text-to-voice and nine new natural voices for GPT-Live
API customers can create reusable custom voices for GPT-Live sessions by providing a voice description in text. Nine natural voices have also been added for German, French, Hindi, Japanese and Korean. Custom voice creation is free; speech generation is billed at the standard GPT-Live rates.
Usage details
Who can use it: All API customers; voice generation is billed at standard GPT-Live session rates.
How to use it: Use the Audio voices endpoint or Voice Playground to create a voice from a text description and connect it to a GPT-Live session.
Tip: Describe the desired pace, age, emotion and pronunciation using brief, concrete wording.
Qwen3.8-Omni-Flash API expands its regional coverage
Model Studio documentation lists China (Beijing), Singapore, Hong Kong, Tokyo, Frankfurt, and Virginia as supported regions for Qwen3.8-Omni-Flash. This clarifies the broader API rollout of the 1M-context model that understands audio and video, including in Europe and the US.
Usage details
Who can use it: Model Studio API users; supported regions include Germany (Frankfurt) and US (Virginia). No separate region is specified for Türkiye.
How to use it: Connect to the Chat Completions or Responses API using the workspace endpoint for your selected region and the API key for that region.
Tip: When using the service from Türkiye, compare Frankfurt with other supported regions based on latency and data residency requirements.
✓ Verified by official sourceModels: qwen3.8-omni-flashSource: Alibaba Cloud Model Studio: Qwen-OmniAuto mode model and idle compaction setting now visible
Claude Code 2.1.302 adds the CLAUDE_CODE_AUTO_MODE_MODEL variable for selecting the model used by the client-side Auto mode safety classifier, and an Idle compaction entry in /config. This lets you manage the safety classifier and conversation compaction separately.
Usage details
Who can use it: Claude Code 2.1.302 and later.
How to use it: Use CLAUDE_CODE_AUTO_MODE_MODEL for Auto mode; turn idle compaction on or off under /config > Idle compaction.
Tip: When changing the classifier model, compare safety refusal and latency metrics on the same project.
✓ Verified by official sourceSource: Claude Code changelogFine-tuning and custom sample operations have been discontinued
Starting October 16, the platform no longer supports creating new fine-tuning tasks or datasets, or creating, modifying, or deleting custom samples; existing information can still be viewed. Fine-tuning and deployment services are scheduled to shut down completely on November 16, 2026.
Usage details
Who can use it: Existing records can be viewed. Users with a previously signed cloud-private service agreement are exempt under the agreement's terms; other users can no longer start new operations.
How to use it: Do not try to create new tasks, datasets, or custom samples. To export existing data or make another request, open a support ticket at https://open.bigmodel.cn/ticket-submit and request the necessary data before shutdown.
Tip: Export custom models and datasets before the November 16 shutdown; users with a cloud-private agreement should confirm migration terms with the support team.
✓ Verified by official sourceSource: Model fine-tuning hizmeti kapanış duyurusuSource: Model deployment hizmeti kapanış duyurusuFork subagents are now the default across most Claude Code surfaces
Fork subagent usage is now the default in claude -p, Agent SDK, Claude Desktop, and VS Code, aligning these surfaces with the existing default behavior in the terminal. Subagents in the main conversation have also been configured to run in the background on desktop and VS Code.
Usage details
Who can use it: Claude Code 2.1.302 and later; terminal, Agent SDK, desktop, and VS Code surfaces.
How to use it: Set CLAUDE_CODE_FORK_SUBAGENT=0 to disable the default; no action is needed to enable it.
Tip: Separate task boundaries and working branches to prevent write conflicts in shared files during parallel subtasks.
✓ Verified by official sourceSource: Claude Code changelogStandard API and Coding Plan keys are now separated
Starting October 16, API keys for the standard model API and GLM Coding Plan are managed through separate entry points. The new key's type determines billing; old keys were classified automatically, their values did not change, and no mandatory reconfiguration was required.
Usage details
Who can use it: Pay-as-you-go or resource package users should use Standard API Key. For an individual Coding Plan, use API Keys → Personal Coding Plan; for a team plan, use the Team Coding Plan → My Plan key.
How to use it: For standard usage, create a Standard API Key at https://z.ai/manage-apikey/apikey-list. For an individual plan, use the Personal Coding Plan tab on the same page; for a team plan, use Team Coding Plan → My Plan. The standard endpoint is https://api.z.ai/api/paas/v4, and the Coding Plan endpoint is https://api.z.ai/api/coding/paas/v4.
Tip: To avoid charges to your account while you have Coding Plan quota, check the key type and base URL together; Standard API Key does not use plan quota.
✓ Verified by official sourceSource: API anahtarı yönetimi kılavuzuClaude Opus 5.5 Fast mode is generally available
Fast mode for Claude Opus 5.5 has exited research preview and is now available without a beta header. The Build and Scale tiers have default Fast mode rate limits; it is also available in Microsoft Foundry deployments on Azure.
Usage details
Who can use it: Claude Opus 5.5 users; Claude Platform and Microsoft Foundry Azure deployments. Preview status continues for Claude Opus 5 and 4.8.
How to use it: Enable Fast mode in a Claude Opus 5.5 request; the fast-mode-2026-02-01 header is no longer required.
Tip: Use the speed boost for non-critical, latency-sensitive tasks; monitor the Fast mode rate limit during periods of heavy traffic.
✓ Verified by official sourceModels: Claude Opus 5.5Source: Claude Platform API sürüm notlarıSource: Claude Opus 5.5 duyurusuCode Review is generally available
Claude Code Code Review has exited research preview and is now generally available for Team and Enterprise organizations. Reviews are about three times faster, and administrators can choose a Quick, Standard, or Deep level for each repository.
Usage details
Who can use it: Claude Team and Enterprise; an administrator must enable it in Claude Code settings and install the GitHub App.
How to use it: Enable Code Review in Claude Code settings, install the GitHub App, and select the repositories to review; reviews will start automatically on new PRs.
Tip: Start with Deep for critical repositories and Quick for small, frequently changing repositories, then monitor review times.
✓ Verified by official sourceSource: Code Review genel kullanıma açıldıGPT-Live-1.1, Quip, and new regional voices arrive
ChatGPT Voice is beginning to switch to GPT-Live-1.1 and its mini version, GPT-Live-1.1 Quip, for more natural accents and a greater variety of brief notifications during searches. New voice options for French, German, Hindi, Japanese, Korean, and Mexican Spanish have been added, along with English Bubbie.
Usage details
Who can use it: Rolling out in ChatGPT Voice; the voice list varies by region and account.
How to use it: Select an available voice at Settings > Voice > Voice.
Tip: If accent accuracy is important, set the Voice language at Settings > Voice > Language before you start speaking.
Muse Spark Open weights released
Meta has released Muse Spark Open weights, with 816 billion total parameters and approximately 42 billion active parameters per token, and a 1 million-token context window. The model is designed for long-horizon agentic tasks, coding and multi-step reasoning, and supports running on your own hardware as well as several hosted API options.
Usage details
Who can use it: The weights are available through Hugging Face, Amazon SageMaker AI, Google Cloud Model Garden and Microsoft Foundry Models; hosted inference is available through Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI. The Muse Model License applies.
How to use it: Download the model from Hugging Face or run it with one of the supported providers; SGLang, vLLM, TensorRT-LLM, Transformers, MLX, llama.cpp and Unsloth are supported.
Tip: For local use, first choose the NVFP4, BF16 or GGUF version suited to your hardware; a 1 million-token context is practical only with enough memory.
✓ Verified by official sourceModels: Muse Spark OpenSource: Introducing Muse Spark Open, The Most Capable US Open-Weight Model to DateSource: Build with Muse Spark OpenQwen Code makes Claude Code and Codex teammates in the same conversation
With Agent Collaboration, you can call Claude Code or Codex into a conversation with @ to handle a task; results, tool steps, and token costs are visible in the same flow. A Qwen Code server on another machine can also be made a teammate, and the agent team can break work into smaller tasks.
Usage details
Who can use it: Experimental feature, disabled by default; macOS/Linux and compatible agent installations.
How to use it: Turn on Enable Agent Collaboration in Settings and restart; create a teammate on the Agents page and call it with @ in a conversation.
Tip: Assign one agent to write code and another to review it so you can track changes and validation in the same conversation.
Multiple Qwen Code instances on one Web Shell page
All Qwen Code instances saved by the user remain connected on the same Web Shell page. Each instance appears in its own session and workspace group, and you can switch between groups without refreshing the page.
Usage details
Who can use it: Web Shell users running multiple local or remote Qwen Code services.
How to use it: Start Qwen Code servers with --allow-origin and add the instances to Web Shell.
Tip: Leave a long build running on a separate machine and track its progress from your laptop on the same page.
Compliance API returns user instructions
The Compliance API now provides, in beta, the Instructions for Claude text saved by Enterprise members in claude.ai and the Global instructions content in Cowork. This lets compliance teams review user preferences alongside organizational records.
Usage details
Who can use it: Claude Enterprise; requires an existing Compliance Access Key and the read:compliance_user_data scope.
How to use it: Call the GET /v1/compliance/organizations/{org_uuid}/user_preferences endpoint.
Tip: When exporting user instructions, apply your access and retention policies for personal or sensitive content.
✓ Verified by official sourceSource: Claude Platform API sürüm notlarıImages are now supported in conversations
Kimi Work 3.2.17 now supports displaying images directly in conversation content.
Usage details
Who can use it: Kimi Work 3.2.17 users.
How to use it: Add an image to a conversation or run a task that includes an image; it will appear in the conversation stream.
Tip: For image-based reviews, check that the image loaded correctly in the conversation stream before asking for a description.
✓ Verified by official sourceSource: Kimi Work Release NotesMulti-turn and deep web search added to MiMo Chat
MiMo Studio V3.4.1 made multi-turn, deeper internet search available in MiMo Chat. The feature aims to improve result accuracy by providing more context for the model’s reasoning.
Usage details
Who can use it: MiMo Chat users in MiMo Studio; the announcement does not mention any separate plan or regional restrictions.
How to use it: Open MiMo Studio and go to MiMo Chat, then start your research request with an initial question; follow up in the same conversation with additional questions and search constraints. The official changelog does not specify a separate activation setting.
Tip: State the scope, date range, and source expectations in your first question; in later turns, ask for comparisons or verification to preserve the search context.
✓ Verified by official sourceSource: Xiaomi MiMo Studio özellik günlüğü (V3.4.1)Models API reports between_tools thinking capability
GET /v1/models and model detail endpoints now report the capabilities.thinking.types.between_tools field. This capability identifies models that can disable initial thinking while preserving ongoing thinking between tool calls.
Usage details
Who can use it: Claude Models API users; the field returns true or false depending on the supported model.
How to use it: Read capabilities.thinking.types.between_tools in the GET /v1/models response and choose the thinking setting accordingly.
Tip: Read this capability field to select compatible models dynamically instead of hard-coding model IDs.
Fixed limit on creating Sites removed for Plus and Pro
Plus and Pro users are no longer subject to a fixed limit on the number of Sites they can create. Usage and storage quotas remain in effect; the change mainly removes the fixed cap on the number of projects.
Usage details
Who can use it: ChatGPT Plus and Pro; usage and storage limits still apply.
How to use it: Create a new Site from the Sites area in ChatGPT; no additional setting is required.
Tip: When creating many Sites, use a consistent naming convention for folders or workspaces to keep content and permissions organized.
✓ Verified by official sourceSource: ChatGPT yayın notlarıQwen Code can disable selected MCP servers at startup
The new --disable-mcp-server option lets you skip a specific MCP server or a list of servers for a single run without changing the settings file. This lets you isolate a problematic or unnecessary service without affecting the overall configuration.
Usage details
Who can use it: Qwen Code CLI users.
How to use it: Run qwen --disable-mcp-server github or, using a comma-separated list, qwen --disable-mcp-server github,postgres.
Tip: When debugging, first disable only the suspected MCP server to narrow down the source of the issue.
Agent Swarm modes can now be selected through Remote Control
Work 3.2.17 improved the remote-control experience, allowing users to select and switch between Agent Swarm modes.
Usage details
Who can use it: Users of Kimi Work Remote Control.
How to use it: Use the Agent Swarm mode selector in a Remote Control session on your phone or remote device.
Tip: Inform users about the expected increase in credit usage and task duration before switching to parallel tasks.
✓ Verified by official sourceSource: Kimi Work Release NotesApplications open for Rosalind Academic Plan
Eligible academic labs conducting life sciences research can apply for Rosalind Academic Plan. Approved university labs in the US can receive up to 30 ChatGPT Business Standard seats free of charge for up to 12 months; access to GPT-Rosalind depends on approval for trusted access.
Usage details
Who can use it: Eligible life sciences labs at US universities.
How to use it: Complete the Rosalind lab application form as the principal investigator or lab manager.
Tip: Before applying, clarify the lab members and the research use case that requires GPT-Rosalind.
✓ Verified by official sourceModels: GPT-RosalindSource: ChatGPT yayın notlarıSource: Rosalind Academic Plan duyurusuGPT-5.5 retired from ChatGPT, Work, and Codex
GPT-5.5 has been removed from ChatGPT, ChatGPT Work, and Codex. Its retirement does not affect API access; developers can continue to access it through the API and manage migration requirements separately.
Usage details
Who can use it: No longer available in ChatGPT, Work, or Codex; API access is unchanged.
How to use it: Select a new model in ChatGPT or Codex; API users should check model status in their platform settings.
Tip: Reassess production workflows pinned to the old model with the new GPT-6 family, and rerun output tests.
Grok Voice 2.0 begins to be deprecated
`grok-voice-2.0` was marked as deprecated, and Grok Voice 3.0 was recommended with the same pricing and API shape as the new version. Existing sessions using `grok-voice-latest` were automatically upgraded to 3.0.
Usage details
Who can use it: Affects API customers using the older versioned slug; new applications should use 3.0 or `latest`.
How to use it: Change the `model` field in requests to `grok-voice-3.0` or `grok-voice-latest`.
Tip: If you use a versioned model in production, validate the migration in a test environment; `latest` makes automatic upgrades easier but does not provide version pinning.
✓ Verified by official sourceModels: grok-voice-2.0, grok-voice-3.0Source: Release NotesSource: Speech to SpeechGrok Voice 3.0 launches
Grok Voice 3.0 is now available in the API as a low-latency speech-to-speech model and, according to xAI, outperforms the previous version in customer support calls and two-way conversations. The model also powers voice interactions at Tesla and SpaceX, as well as Grok Bot’s voice mode.
Usage details
Who can use it: Available to Speech to Speech API users; priced at 0,08 US dollars per minute, and `grok-voice-latest` routes to this model automatically.
How to use it: Use `model=grok-voice-3.0` or `model=grok-voice-latest` in a WebSocket Speech to Speech connection; existing API parameters remain unchanged.
Tip: Start with `grok-voice-latest` for low-latency customer service scenarios; pin the versioned slug if you don’t want version changes.
✓ Verified by official sourceModels: grok-voice-3.0Source: Introducing Grok Voice 3.0Source: Release NotesLinux client, Remote Control, and File Tree released
Kimi Code Desktop 1.1.0 became available for Linux as AppImage, deb, and RPM packages; the same release added local Agent control from another device and workspace file browsing in the right sidebar.
Usage details
Who can use it: Linux, macOS, and Windows Desktop users; Remote Control is enabled in settings.
How to use it: Install the Linux package; enable Remote Control in Settings; open the File Tree panel from the right sidebar.
Tip: When working on a server or remote Linux machine, restrict the Remote Control connection securely using a QR code or link.
✓ Verified by official sourceSource: Kimi Code Desktop ChangelogOpenTelemetry export launches in beta
OpenTelemetry export is now available in beta for sending Messages API requests and Claude Managed Agents sessions to observability platforms. The Python, TypeScript, C#, Go, Java, PHP, and Ruby SDKs support span generation.
Usage details
Who can use it: Claude Console and Claude Managed Agents users; beta.
How to use it: Add an OpenTelemetry destination in Claude Console and enable span settings in the SDK you use.
Tip: Mask prompt content in traces and store only latency, token, and tool-call metrics by default.
✓ Verified by official sourceSource: Claude Platform API sürüm notlarıV2.5 API requests now automatically route to V2.6 models
As of October 14, 2026 18:00 (UTC+8), `mimo-v2.5-pro` requests are automatically routed to `mimo-v2.6-pro`, and `mimo-v2.5` requests to `mimo-v2.6-flash`. V2.5-Pro routing enables multimodal understanding; in thinking mode, V2.6 models do not accept custom `temperature` and `top_p` values and effectively use 1.0 and 0.95.
Usage details
Who can use it: Users sending requests to the Xiaomi MiMo Open Platform API using the old V2.5 model names are affected; requests are processed as the new models and billed at the new model rates.
How to use it: Switch to the new model IDs in your integration. If you continue using the old names until the migration is complete, send `max_completion_tokens` explicitly and remove `temperature`/`top_p` customization in thinking mode.
Tip: Monitor the actual model ID and usage cost in request logs; because automatic routing silently switches to the new model, recheck output-length and budget assumptions.
✓ Verified by official sourceModels: mimo-v2.5-pro, mimo-v2.5, mimo-v2.6-pro, mimo-v2.6-flashSource: MiMo model emeklilik günlüğüSource: MiMo model emeklilik günlüğü (Çince)Experimental interim progress updates added
CLI 2.2.0 introduced a feature, disabled by default, that lets the Agent send brief NotifyUser progress messages while a task is in progress.
Usage details
Who can use it: Kimi Code CLI users; the experimental setting must be enabled in new sessions.
How to use it: Enable NotifyUser from the /experiments panel in the TUI.
Tip: Enable it to see interim status during long tasks; leave it off to avoid unnecessary noise in sessions that require short responses.
✓ Verified by official sourceSource: Kimi Code CLI ChangelogReal-time progress view added to Desktop conversations
Desktop 1.1.0 introduced experimental progress updates, disabled by default, that show the Agent’s real-time work progress within a conversation.
Usage details
Who can use it: Kimi Code Desktop 1.1.0 users.
How to use it: Turn on the progress updates option under Settings → Lab.
Tip: Enable it to improve visibility during long coding tasks; turn it off when you want to keep the screen uncluttered.
✓ Verified by official sourceSource: Kimi Code Desktop ChangelogFile and skill usage limits increased
In Grok Bot, the limit for a single file attachment increased to 250 MB, the limit for a video attachment to 1 GB, and the number of attachments per message to 20. The number of skills bots can use from installed plugins also increased from 24 to 200.
Usage details
Who can use it: Mac and Windows users with Grok Bot access; the 200 MB total message limit and plan limits still apply separately.
How to use it: Attach files to a Bot chat as usual; skills are selected automatically from installed plugins.
Tip: Before sending large videos, check the 1 GB limit and the requirement that the Bot be able to view the video.
✓ Verified by official sourceSource: Grok Bot ChangelogAccessible art tour integration for Pinacoteca di Brera
Meta, Pinacoteca di Brera and EssilorLuxottica have launched a pilot combining AI glasses with the DescriVedendo audio tour in the Grande Brera app for blind and low-vision visitors. The glasses recognize eight selected artworks and play their audio descriptions; Meta AI answers questions about the surroundings, and Be My Eyes provides real-time human assistance.
Usage details
Who can use it: By reservation at Pinacoteca di Brera in Milan, Italy; AI glasses are available as part of the pilot Tuesday through Friday, 10:00 a.m.–4:30 p.m.
How to use it: Reserve the Grande Brera tour and enable artwork recognition on the glasses; choose Meta AI or Be My Eyes for additional help.
Tip: Use open-ear headphones for museum directions and artwork descriptions so you can continue hearing sounds around you.
✓ Verified by official sourceSource: Meta, Milan’s Pinacoteca di Brera, and EssilorLuxottica Expand Access to Art Through AI Glasseskimi app command opens a new Desktop chat from the terminal
Kimi Code CLI 2.2.0 added the kimi app [path] command, which opens a new Kimi Code Desktop draft in the current or specified folder.
Usage details
Who can use it: Developers using Kimi Code CLI and Desktop.
How to use it: Run kimi app or kimi app /project/path in the terminal.
Tip: After exploring with the CLI, specify the folder path to move the same workspace into the GUI.
✓ Verified by official sourceSource: Kimi Code CLI ChangelogScheduled Managed Agents deployments come to AWS
Managed Agents scheduled deployments are now available on Claude Platform on AWS. Scheduled agent runs can be deployed within a Platform installation hosted on AWS.
Usage details
Who can use it: Claude Platform on AWS users.
How to use it: Create a scheduled deployment for Managed Agents and select the AWS Platform environment as the target.
Tip: Use idempotent operations and safeguards against duplicate runs for scheduled tasks.
✓ Verified by official sourceSource: Claude Platform API sürüm notlarıAntigravity coding agent embedded in the Gemini macOS app
The Antigravity agent in the Gemini macOS app can now start coding, web experiences, and prototyping directly from a chat. After its plan is approved, the agent creates and tests the code in the background and notifies the user when it is ready.
Usage details
Who can use it: English; personal Google accounts for users aged 18 and over and eligible work accounts; regions where Antigravity is supported.
How to use it: On a Mac, click Antigravity at the top of the Gemini app, approve the agent's plan, and bring the result back into the Gemini chat.
Tip: Limit the agent's write and execution permissions to only the project folder it needs.
✓ Verified by official sourceModels: Google AntigravitySource: Gemini Apps' release updates & improvementsGemini 4 Argon reaches GA in the API
Gemini 4 Argon is generally available in the Gemini API for complex coding, agent, math, and science tasks. The model offers a 1 million-token combined context/output window, long decoding, and low/medium/high thinking_level options.
Usage details
Who can use it: Paid Gemini API tier and Google AI Studio; not available on the free API tier.
How to use it: Set the model field to gemini-4-argon; for difficult tasks, use generation_config.thinking_level, and for long outputs, use background=true with max_output_tokens.
Tip: Remove temperature, top_p, and top_k from existing integrations, then reassess the quality-cost balance using thinking_level.
✓ Verified by official sourceModels: Gemini 4 ArgonSource: Release notes | Gemini API - Google AI for DevelopersSource: What's new in Gemini 4 ArgonIntroductory API pricing announced for Gemini 4 Argon
The Gemini 4 Argon API is priced at $2 per 1 million input tokens and $10 per 1 million output tokens through November 30, 2026. Starting December 1, prices will rise to $4 and $20, respectively; cached requests receive a 95% discount on the input price.
Usage details
Who can use it: Paid Gemini API tier; Gemini 4 Argon is not available on the free tier.
How to use it: Use the gemini-4-argon model on the paid Gemini API tier; include thinking tokens in cost estimates.
Tip: First measure cost and quality with low or medium thinking_level, and use high only for difficult tasks.
✓ Verified by official sourceModels: Gemini 4 ArgonSource: What's new in Gemini 4 ArgonSource: Gemini 4 Argon is now available for developers.Gemini 4 Argon becomes available in the app
Gemini 4 Argon has arrived in the Gemini app with stronger reasoning for complex tasks, more direct answers, and better progress updates on longer tasks. The same day, Google also announced the ability to switch from the main chat to task mode and broader support for connected apps.
Usage details
Who can use it: Google AI Pro and Google AI Ultra subscribers; country, age, and language restrictions apply to some features.
How to use it: Select Gemini 4 Argon from the model picker in the Gemini app; switch the chat to Tasks mode for eligible tasks.
Tip: For longer tasks, check interim outputs and progress updates to correct misdirection early.
✓ Verified by official sourceModels: Gemini 4 ArgonSource: Gemini 4 Argon comes to the Gemini app.Source: Gemini Apps' release updates & improvementsSkills rollout begins in the app, Workspace rollout delayed
Google said the Skills feature in the Gemini app began a gradual rollout on October 13 and is expected to be fully rolled out by mid-November. The rollout of Skills in Workspace has been postponed to a later date due to security and reliability work.
Usage details
Who can use it: Personal and Workspace accounts in the Gemini app; a new date for Workspace Skills has not yet been announced, and age and license requirements apply.
How to use it: Use the Skills help flow in the Gemini app; Workspace administrators should wait for a new rollout announcement for Skills.
Tip: Teams using Gems should back up their existing Gems workflows during the transition and test Skills equivalents separately.
✓ Verified by official sourceSource: An update on the launch of skills in the Gemini app and WorkspaceYour algorithm lets users tune Feed and Reels recommendations with text
Facebook has begun rolling out the AI-powered Your algorithm section, where users can write in their own words what topics they want to see in Feed and Reels. Interested and Not interested feedback can also be refined, and the ranking system updates recommendations in real time within the same session.
Usage details
Who can use it: Rolling out gradually to Facebook Feed and Reels users; it may not appear, depending on the account and region.
How to use it: Open the three-dot menu on a post in Feed or Reels and go to Your algorithm; add topics or give Interested/Not interested feedback.
Tip: Write measurable preferences instead of general statements, such as 'less local politics, more Turkish-language AI training.'
✓ Verified by official sourceSource: Your Algorithm, Your Words: New Ways to Shape What You See on FacebookX Subscription combines Grok, Grok Bot, Cursor, and X
xAI launched The X Subscription, combining Grok, Grok Bot, Cursor, and X Premium benefits under one subscription. The Free, Plus, Premium, Super, and Ultra tiers were announced at 0, 8, 30, 100, and 200 US dollars/month, respectively; usage of Grok and Cursor in paid tiers draws from a shared pool.
Usage details
Who can use it: Available immediately to new customers; being rolled out gradually to existing SuperGrok, Cursor, and X Premium subscribers, with global availability targeted for the days following the announcement.
How to use it: Compare plans at x.ai/pricing; existing subscribers can see the option to switch and, if applicable, a prorated refund offer under Billing settings.
Tip: If you use Grok and Cursor heavily together, check the shared usage pool and note that third-party model usage is billed separately.
✓ Verified by official sourceSource: Introducing The X SubscriptionMaximum Usage plan added for Muse Code
Meta has added a new Maximum Usage plan to Muse Code, offering 80 times the Everyday Usage limit. Alongside Power Usage, the plan is one of the subscription options that provides access to Muse Wonder Full, and retains the five-hour and weekly reset cycles.
Usage details
Who can use it: For Muse Code subscribers; Maximum Usage and Power Usage plans include access to Muse Wonder Full. The announcement did not specify the price.
How to use it: Select the Maximum Usage or Power Usage plan on the Muse Code subscription page; choose Muse Wonder Full from the model selector.
Tip: Use the 80-times limit for extended, multi-file coding tasks; a lower plan may be enough for short tasks.
✓ Verified by official sourceModels: Muse Wonder FullSource: Put your ideas to work with Muse Wonder and Muse Wonder FullMuse Wonder and Muse Wonder Full released
Meta introduced the Muse Wonder and Muse Wonder Full models for personal and professional agentic tasks. The models carry important context across conversations with Active Memory, use Active Focus to try alternative approaches and recover from errors on difficult tasks, and can work in areas such as coding, computer use, law, finance and healthcare.
Usage details
Who can use it: Muse Wonder is available in Meta Model API and Muse Code; API pricing is $4 per 1 million input tokens and $20 per 1 million output tokens. Muse Wonder Full is available only with a Muse Code subscription, on the Power Usage and Maximum Usage plans.
How to use it: Use muse-wonder-1.0 as the model ID in Meta Model API; in Muse Code, select the model from the relevant subscription plan.
Tip: For long tasks, clearly state the goal, constraints and key approval points in the first prompt; make use of Active Focus generating alternative plans.
✓ Verified by official sourceModels: Muse Wonder, Muse Wonder FullSource: Introducing Muse Wonder and Muse Wonder Full: Frontier Models for AgentsSource: Put your ideas to work with Muse Wonder and Muse Wonder FullQwen-Audio-3.1-TTS-Flash arrives for real-time synthesis
Qwen-Audio-3.1-TTS-Flash is a new version aimed at low-latency interactions through streaming speech synthesis and system audio. It offers a more suitable Flash option for voice assistants, real-time dialogue, and customer service.
Usage details
Who can use it: Model Studio International API users.
How to use it: Select the qwen-audio-3.1-tts-flash model in the API call and play the streaming response chunk by chunk.
Tip: Reduce perceived latency in voice assistants by starting playback without waiting for the first audio chunk to finish.
Background tasks no longer interrupt browser preview
Work 3.2.16 ensures the built-in browser is not interrupted while tasks run in the background.
Usage details
Who can use it: Kimi Work users.
How to use it: Start a task while a page is open in the built-in browser, then send the app to the background.
Tip: Keep the browser tab open when doing web research and generating files at the same time.
✓ Verified by official sourceSource: Kimi Work Release NotesGLM-Legal opens beta access for legal workflows
Z.ai introduced GLM-Legal, built on GLM-5.3-Flash with legal-domain mid-training and reinforcement-learning post-training. The model focuses on multistep legal work such as contract review, evidence analysis, compliance research, and document drafting, and is accepting applications for beta access.
Usage details
Who can use it: Law firms, legal technology companies, in-house legal teams, and independent practitioners were invited to the beta program. API access is offered to selected users by application, while AutoClaw trials are available through the channel specified in the announcement; no public model identifier or open weights were announced.
How to use it: Apply for API beta access at https://z.ai/consultation/legal. For a ready-made workflow, open AutoClaw, provide a contract, evidence, or compliance task, and attach the source documents.
Tip: Clearly identify legal sources and documents when providing them to the model; have an authorized legal professional review the results.
✓ Verified by official sourceModels: GLM-Legal, GLM-5.3-FlashSource: GLM-Legal: Ajan tabanlı hukuk zekâsı (resmî duyuru)Source: AutoClaw'da GLM-Legal beta duyurusuPDF preview engine upgraded with zoom and quote selection controls
Kimi Work 3.2.16 upgraded the PDF preview engine, improving zoom, quote controls, and text selection interactions.
Usage details
Who can use it: Kimi Work 3.2.16 users.
How to use it: Open a PDF in Work preview and use the zoom, text selection, or quote controls.
Tip: When preparing a sourced report, select the quote bubble directly and link it to the relevant section.
✓ Verified by official sourceSource: Kimi Work Release NotesBot gets an email address and can search X without an account
Bots can now get their own email addresses and receive mail, and can search X posts and profiles without connecting an X account. In addition, apps such as Figma, Notion, and Canva can display interactive cards in chat.
Usage details
Who can use it: For Grok Bot users; some email and connector capabilities depend on plan or administrator policies.
How to use it: Request an email address or ask for an X search from within Grok Bot; enable app cards through Connect Apps.
Tip: Use date, like, and repost filters in X search to narrow large result sets.
✓ Verified by official sourceSource: Grok Bot ChangelogPayment tools and QuickBooks Online connector added
Grok Bot can make payments with X Money or Stripe Link; QuickBooks Online was also added to the list of connectable apps. This change expands the Bot’s ability to initiate approved transactions in financial workflows, rather than only providing information.
Usage details
Who can use it: Users with Grok Bot access; payment and finance connectors are subject to account, region, and approval policies.
How to use it: Connect QuickBooks Online from Connect Apps; during payment, select X Money or Stripe Link from the approval card.
Tip: In payment flows, use Auto-review to require approval for every purchase and money transfer step.
✓ Verified by official sourceSource: Grok Bot ChangelogPlanned model retirements take effect in Model Studio
Alibaba Cloud retired multiple older or snapshot model groups on October 10, 2026. After retirement, the affected API calls fail; users need to find the affected models through Model Monitoring and test their replacements.
Usage details
Who can use it: Affected Model Studio users; accounts using the relevant models in the China (Beijing) and Singapore regions.
How to use it: Check the retired models on the Model Studio > Model Monitoring page and migrate production model IDs to current versions.
Tip: To avoid confusion, track fixed-date or currently supported model IDs in production and monitor retirement notices.
✓ Verified by official sourceSource: Model decommissioning policySource: Model Studio release notes and feature updatesAutoClaw offers weekly rewards of up to 132M GLM-5.3-Flash tokens
During its October 13–16 campaign, AutoClaw let all users claim 8 million GLM-5.3-Flash tokens (800 credits) each day. Eligible paid members could receive an additional daily reward based on their active membership tier; rewards expired at midnight UTC+8 on the day they were claimed.
Usage details
Who can use it: All AutoClaw users were eligible for the base reward. AutoClaw monthly or group subscribers and GLM Coding Plan individual Lite, Pro, and Max members were eligible for an additional reward; multiple memberships could not be combined.
How to use it: Users had to log in to AutoClaw on campaign days and claim the daily reward; each account could claim once per day. The reward had to be used by midnight UTC+8 that same day.
Tip: Claim the daily reward right after logging in and use it before starting long-running tasks; it did not carry over to the next day.
✓ Verified by official sourceModels: GLM-5.3-FlashSource: AutoClaw haftalık ayrıcalıkları: 132M token ve üye ödülleriAgents API sandbox prewarming and active-turn billing
OpenAI-hosted sandboxes can be prewarmed before a user message is completed, reducing session startup latency. Sandbox computation is now billed during active turns rather than during idle time between turns.
Usage details
Who can use it: Developers using OpenAI-hosted sandboxes in the Agents API.
How to use it: Enable pre-warm configuration for the sandbox before connecting it to the agent session.
Tip: Test prewarming for independent turns that require long waits, and keep it off for short tasks to avoid unnecessary resource consumption.
✓ Verified by official sourceSource: OpenAI API changelogManaged Agents dynamic workflows enter beta
Managed Agents agents can now write a dynamic workflow that runs in the background for multi-part tasks and run multiple agents in stages. The feature aims to scale tasks such as reviewing hundreds of documents by combining their results.
Usage details
Who can use it: Beta; requires the managed-agents-2026-04-01 beta header and a workflows enabled configuration in the multiagent field.
How to use it: Set multiagent: {type: multiagent_20261001, workflows: {type: enabled}} in the agent definition and monitor workflow_run events.
Tip: Test the workflow on a small set of documents and verify stage outputs and reruns of failed agents.
✓ Verified by official sourceSource: Claude Platform API sürüm notlarıAutomatic migration and retirement schedule announced for MiMo V2.5 models
Xiaomi announced that as of October 14, 2026 18:00 (UTC+8), `mimo-v2.5-pro` and `mimo-v2.5` requests will be automatically routed to the `mimo-v2.6-pro` and `mimo-v2.6-flash` models, respectively, and that the old model names will become invalid on October 21, 2026 10:00. In the migration, moving from V2.5-Pro to V2.6-Pro enables multimodal understanding; moving from V2.5 to V2.6-Flash raises the default `max_completion_tokens` value from 32768 to 131072.
Usage details
Who can use it: All Xiaomi MiMo Open Platform API integrations using the old model IDs are affected; routed requests are billed at the new model rates.
How to use it: Check pending retirements in the Bill Details/Usage section of the MiMo console. In code, use `mimo-v2.6-pro` instead of `mimo-v2.5-pro` and `mimo-v2.6-flash` instead of `mimo-v2.5`; test the behavior of `temperature`, `top_p`, and `max_completion_tokens`.
Tip: Specify `max_completion_tokens` explicitly instead of relying on defaults, and run regression tests with both text and multimodal requests before migrating.
✓ Verified by official sourceModels: mimo-v2.5-pro, mimo-v2.5, mimo-v2.6-pro, mimo-v2.6-flashSource: MiMo model emeklilik günlüğüSource: MiMo model emeklilik günlüğü (Çince)Qwen Code’s /batch-api runs background jobs for half the real-time price
The new /batch-api command first schedules independent, repetitive tasks for free, then runs them in the background after approval. According to the documentation, batch jobs are billed at half the real-time rate, and results are delivered later as files.
Usage details
Who can use it: Qwen Code and an eligible Bailian/DashScope account; charges apply to the API account.
How to use it: Run the /batch-api command with a task description, approve the sampling plan, and wait for the result files.
Tip: Send hundreds of independent documents to batch instead of waiting for them one by one; use a regular conversation for dependent tasks.
✓ Verified by official sourceSource: Qwen Code Weekly: Qwen Live Is Live, Work the Pages You're Already Logged Into, Assign Tasks by EmailQwen Code accepts remote tasks through Email Channel
Email Channel sends an email from a phone to Qwen Code on a computer as a task, then returns the result by email. The user’s address is added to an allowlist so that only authorized senders can start tasks.
Usage details
Who can use it: Qwen Code Web Shell; mailbox and allowlist configuration required.
How to use it: Connect a mailbox in Web Shell and add your email address to the allowlist.
Tip: For initial testing, use the email channel only for reversible tasks, such as a harmless log summary.
✓ Verified by official sourceSource: Qwen Code Weekly: Qwen Live Is Live, Work the Pages You're Already Logged Into, Assign Tasks by EmailQwen Code handles browser tasks in existing Chrome sessions
With the Chrome extension, Qwen Code can navigate pages, fill in forms, and submit them in the background using the user’s open tabs and existing sign-ins. Each session gets its own tab group; Windows is not yet supported.
Usage details
Who can use it: Mac and Linux with the Chrome extension; Windows is not supported.
How to use it: Install the Qwen Code extension from the Chrome Web Store and request a browser task in a conversation.
Tip: First verify the correct tab and account scope with a read-only summarization task, then allow form submission.
✓ Verified by official sourceSource: Qwen Code Weekly: Qwen Live Is Live, Work the Pages You're Already Logged Into, Assign Tasks by EmailQwen Live is now available with voice conversations and screen sharing
Qwen Live offers real-time voice conversations in the Web Shell and desktop version; when screen sharing is enabled during a call, Qwen can see errors on the screen. The feature is enabled with a Bailian DashScope Realtime API key.
Usage details
Who can use it: Qwen Code Web Shell and desktop; a Bailian DashScope Realtime API key is required.
How to use it: Open Live in the left menu, paste your DashScope Realtime API key and click Save; then use Open Live Voice and screen sharing.
Tip: When sharing an error window, make sure personal notifications and secrets aren’t visible.
✓ Verified by official sourceSource: Qwen Code Weekly: Qwen Live Is Live, Work the Pages You're Already Logged Into, Assign Tasks by EmailQwen-Image-2.1-Turbo released
Qwen-Image-2.1-Turbo was introduced as an open-weight option balancing speed, quality, and cost in the Qwen-Image series. It targets fast creative generation with an alpha channel, unified generation and editing, multiple references, and improved typography.
Usage details
Who can use it: Users covered by Model Studio International; the open-weight version can also run on their own infrastructure.
How to use it: Select the qwen-image-2.1-turbo model in Model Studio, or use Qwen’s open-weight repository for local inference.
Tip: When generating transparent PNG assets and product images, try requesting the background as a separate layer.
Model access narrowed for free and AI Plus plans
Starting October 9, users without an AI subscription on personal accounts were limited to Flash-Lite; AI Plus users retained access to Flash-Lite and Flash but lost Pro models. AI Pro and AI Ultra users reportedly saw no reduction in access, and Deep Think access was added to AI Pro.
Usage details
Who can use it: Personal accounts; model access varies by plan. This change, based on Google's support documentation, was reported by reputable technology news outlets.
How to use it: Check the model picker in the Gemini app and your Google AI plan; AI Pro or AI Ultra is required for more powerful Pro models.
Tip: If you rely on critical workflows on the free or Plus plan, compare prompt and output quality before the model change.
✓ Verified by official sourceModels: Gemini 3.5 Flash-Lite, Gemini 3.6 Flash, Gemini 3.1 ProSource: Gemini app limiting what models free & AI Plus users can access, AI Pro adding Deep ThinkSource: Google is changing Gemini AI plans from October 9, here is what it means for youQwen Code’s Trajectory tab shows the source of latency
The Trajectory panel displays model requests, tool calls, and subagent requests on separate timelines. Users can select a specific interval and expand the input, output, and token cost to investigate why a run was slow or expensive.
Usage details
Who can use it: Qwen Code Web Shell users.
How to use it: After a run, open the Trajectory tab in the right panel and click the time interval you want to inspect.
Tip: Identify whether latency comes from the model, a tool, or a subagent, then optimize only the bottleneck layer.
✓ Verified by official sourceSource: Qwen Code Weekly: Qwen Live Is Live, Work the Pages You're Already Logged Into, Assign Tasks by EmailClaude Dashboards and Claude Motion enter beta
Claude Dashboards can now generate refreshable dashboards from connected sources such as BigQuery, Databricks, Snowflake, or Salesforce. Claude Motion creates animated explainers with editable code and can export the output as MP4.
Usage details
Who can use it: Dashboards is in beta on paid plans; Motion is in beta on Team and Enterprise. An Enterprise administrator must enable it in Artifacts settings.
How to use it: Request a dashboard in the chat box or type /motion to open the Motion option.
Tip: Verify the query and last refresh time for every chart in a dashboard before sharing it.
✓ Verified by official sourceSource: Claude ile canlı dashboard ve animasyon oluşturmaDocs, Slides, and Design exit beta
Claude Docs, Claude Slides, and Claude Design are now available without a beta label on all Claude plans, including Free. Artifacts updates were also announced, including CMEK, team editing, external sharing, format-preserving exports, and mobile editing improvements.
Usage details
Who can use it: Free, Pro, Max, Team, and Enterprise; on Enterprise, availability depends on administrator policies, with the default enablement date specified as October 15, 2026.
How to use it: Request Docs, Slides, or Design in any conversation, or choose a template from the Artifacts tab.
Tip: Before sharing team templates, have the Enterprise administrator check Artifacts and sharing policies.
✓ Verified by official sourceSource: Claude ile canlı dashboard ve animasyon oluşturmaSource: Artifacts yardım makalesiLegacy Deep Research agent retired
The deep-research-pro-preview-12-2025 agent was deprecated and shut down on October 23, 2026. Google recommends deep-research-preview-04-2026 for faster jobs and deep-research-max-preview-04-2026 for the most comprehensive automated research.
Usage details
Who can use it: Developers using managed agents in the Gemini API Interactions API.
How to use it: Replace the agent parameter in the interactions.create call with one of the recommended new Deep Research agents.
Tip: During migration, compare report length, source count, cost, and latency using the same test questions.
✓ Verified by official sourceModels: deep-research-pro-preview-12-2025, deep-research-preview-04-2026, deep-research-max-preview-04-2026Source: Release notes | Gemini API - Google AI for DevelopersSource: Gemini deprecationsGemini 3.7 Flash deprecated and redirected to 3.8
Gemini 3.7 Flash can no longer be selected in AI Studio, and API requests are automatically redirected to Gemini 3.8 Flash. Google recommends updating the model string in clients to gemini-3.8-flash.
Usage details
Who can use it: Gemini API and Google AI Studio users; the old endpoint has not been shut down, but automatic redirection is in effect.
How to use it: Change the model string in your code from gemini-3.7-flash to gemini-3.8-flash and reevaluate the outputs.
Tip: Rather than relying on automatic redirection, manage the version migration with controlled testing and a fixed evaluation set.
✓ Verified by official sourceModels: Gemini 3.7 Flash, Gemini 3.8 FlashSource: Release notes | Gemini API - Google AI for DevelopersSource: Gemini deprecationsGoogle Docs, Sheets, and Slides integration enters beta
Claude for Google Workspace lets users access Claude within files through an add-on that appears as a side panel in Google Docs, Sheets, and Slides. Claude also offers three new connectors for creating and editing Google files from its own chat.
Usage details
Who can use it: Beta on all paid Claude plans; in Team and Enterprise, an owner or primary owner must enable the connectors.
How to use it: Install the add-on from Google Workspace Marketplace and use Extensions > Claude > Open Claude; to edit from chat, enable the Google connectors.
Tip: Keep the default Ask before edits mode and review each change in the approval card.
✓ Verified by official sourceSource: Claude artık Google Docs, Sheets ve Slides ile çalışıyorAPI Ultrafast mode for GPT-6.1 Sol
The Ultrafast service tier is now available for GPT-6.1 Sol in the Responses API. `service_tier: ultrafast` reduces the time between output tokens; global processing and US and EU data residency are supported.
Usage details
Who can use it: All API users, subject to rate limits and supported data residency.
How to use it: Send `service_tier: "ultrafast"` with `gpt-6.1-sol` in a Responses request.
Tip: For real-time streams, use Ultrafast only for requests where latency matters more than cost.
Gemini 3.5 Flash deprecated and routed to 3.6
Gemini 3.5 Flash was removed from the AI Studio model picker, and API requests began being automatically routed to Gemini 3.6 Flash. Google recommends that apps update their model string.
Usage details
Who can use it: Gemini API and Google AI Studio users; automatic routing is available for the old endpoint.
How to use it: Use gemini-3.6-flash or, when appropriate, gemini-3.8-flash instead of gemini-3.5-flash in requests.
Tip: Re-measure token consumption, latency, and structured output behavior with the new model.
✓ Verified by official sourceModels: Gemini 3.5 Flash, Gemini 3.6 FlashSource: Release notes | Gemini API - Google AI for DevelopersSource: Gemini deprecationsBots create PowerPoint and Google Slides presentations
Grok Bot can now design a presentation with the user using example slides and deliver the output as a PowerPoint file or Google Slides. The same release also added support for formatted email drafts.
Usage details
Who can use it: Grok Bot users; Google Slides delivery depends on the permissions of the connected Google account.
How to use it: Ask a Bot to prepare a presentation, choose the example-slide approach, and get the final output as PowerPoint or Google Slides.
Tip: Before generating a presentation, specify the target audience, slide count, and brand template clearly; keep final sharing subject to human approval.
✓ Verified by official sourceSource: Grok Bot ChangelogClaude Haiku 5.5 released
Claude Haiku 5.5 launched as Anthropic’s fastest, lowest-cost small model for high-volume, latency-sensitive tasks. It offers a 1M token context window, 128K maximum output, adaptive thinking, and adjustable effort.
Usage details
Who can use it: Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud, and Microsoft Foundry; Türkiye is on the list of supported regions, but no separate rollout for Türkiye was specified for this release.
How to use it: Use the claude-haiku-5-5 model ID on Claude Platform; in Claude apps, select Haiku 5.5 from the model picker.
Tip: Try Haiku 5.5 as a helper model alongside more expensive models for summarization, classification, compaction, and subagent tasks.
✓ Verified by official sourceModels: Claude Haiku 5.5Source: Claude Haiku 5.5 duyurusuSource: Claude Platform API sürüm notlarıSource: Desteklenen bölgeler politikasıIntelligent UI arrives in ChatGPT with GPT-6
GPT-6 brings Intelligent UI to ChatGPT, enabling it to generate not just text responses but also charts, buttons, forms, tables, and interactive mini-experiences. The model can choose the format of text, visuals, and interactive elements based on the question, and can begin showing its response progressively while thinking.
Usage details
Who can use it: Global rolling rollout: Plus, Pro, Business, and Enterprise on October 7; Free and Go plans starting October 8. Enterprise access depends on admin settings.
How to use it: Ask ChatGPT Chat to explain or compare a topic, or to create an interactive tool.
Tip: Use specific prompts such as “interactive comparison,” “calculator,” or “step-by-step visual” to determine the result format.
✓ Verified by official sourceModels: GPT-6 Sol, GPT-6 Luna, GPT-6Source: GPT-6 ve Intelligent UI duyurusuSource: ChatGPT yayın notlarıMonthly API credits for Max, Ultra, and Team subscribers
Claude subscriptions now include monthly API credits to run your own apps and agents on Claude Platform. The announcement lists $100 or $200 for Max, $200 for Ultra, and up to $500 pooled based on seats for Team; credits are separate from plan limits and do not roll over.
Usage details
Who can use it: The official announcement covers Max, Ultra, and Team; the help article details eligibility for Max and Team. Free, Pro, and Enterprise are not eligible.
How to use it: Use claude.ai > Settings > Billing or, for Team, Organization settings > Billing > API credits > Link organization.
Tip: Since you can link the credit to only one Console organization, check your choice before selecting a production or testing organization.
✓ Verified by official sourceSource: Claude Haiku 5.5 duyurusuSource: Max ve Team aylık API kredileriSource: Claude uygulama sürüm notlarıBrowser/computer toolsets added to Python and TypeScript SDKs
The Python and TypeScript SDKs added beta classes for browser use and computer use tools. Developers can write their own browser or desktop automation methods while the SDK handles the tool loop, URL and file policies, and approval callback.
Usage details
Who can use it: Beta for Python and TypeScript SDK users; Haiku 5.5 is especially targeted at fast, high-volume use.
How to use it: Subclass the browser use or computer use toolset class in the SDK and implement your own automation method.
Tip: Before using these with real accounts, verify URL permissions and the human approval callback in a limited test environment.
✓ Verified by official sourceModels: Claude Haiku 5.5Source: Claude Haiku 5.5 duyurusuSource: Claude Platform API sürüm notlarıSonnet 5.5 prompt cache read price cut in half
The cache read price for Claude Sonnet 5.5 was reduced from $0.20 to $0.10 per million tokens. Cache write, input, and output prices are unchanged; total costs for most long agentic tasks are expected to fall by about 20%.
Usage details
Who can use it: Claude API and Claude Platform surfaces with token-based pricing.
How to use it: No changes are needed to existing prompt caching requests; monitor Sonnet 5.5 usage at the new price.
Tip: Get more value from the discount by caching shared system instructions for long tasks.
✓ Verified by official sourceModels: Claude Sonnet 5.5Source: Claude Haiku 5.5 duyurusuSource: Claude Platform API sürüm notlarıchat-latest snapshot updated to ChatGPT’s current model
The `chat-latest` snapshot has been updated to point to the latest model available to Plus, Pro, Business, and Enterprise users in ChatGPT. OpenAI recommends the GPT-6 family for production use while retaining this model for experiments that approximate ChatGPT usage.
Usage details
Who can use it: OpenAI API customers; the snapshot may change regularly alongside model updates in ChatGPT.
How to use it: Use the `chat-latest` model ID in an API request, or switch to one of the fixed GPT-6 model IDs for production.
Tip: If stable production behavior is required, consider a dated or fixed model ID instead of `chat-latest`.
Native iPad app for Muse released
Meta's Muse iOS app now offers native support for the iPad's large screen and iPadOS multitasking features. The same update also highlighted new connectors for business tools such as Canva, Dropbox, Figma, QuickBooks, GitHub, Zoom, Asana, Klaviyo, Granola, and Notion, along with the ability to plan in the background around small-business goals.
Usage details
Who can use it: Muse users on iPadOS; app and some connectors may vary by account, region, and phased rollout.
How to use it: Install the latest version of Muse from the App Store and open the app on your iPad; add connectors in the app settings.
Tip: Use the large screen to review and edit a marketing plan or product page draft prepared by Muse at the same time.
Web tools in restricted network environments now use allowed_hosts
In Managed Agents environments using limited networking, allowed_hosts now also applies to web_search and web_fetch. In addition, web_fetch can retrieve only URLs previously seen in the session context, reducing the risk of agents exporting data.
Usage details
Who can use it: Claude Managed Agents; limited network environments with web tools enabled.
How to use it: Add the required hosts to allowed_hosts in the environment definition, and bring the URL into context through a user message or web_search result before calling web_fetch.
Tip: Test allowed_hosts and allowed_domains together to verify redirect and subdomain behavior.
✓ Verified by official sourceSource: Claude Platform API sürüm notlarıChatGPT now supports audio file uploads
Users can now upload audio files to ChatGPT to generate transcripts, summarize recordings, and ask questions about their content. Meeting, interview, or lecture recordings can be turned into structured notes or draft follow-up emails.
Usage details
Who can use it: Paid ChatGPT subscriptions and workspaces; availability depends on region, client version, workspace settings, and model support.
How to use it: Attach a supported audio file to a conversation and enter your request for a transcript, summary, or Q&A.
Tip: Before relying on key meeting decisions, compare the transcript with the recording; transcription errors may occur.
✓ Verified by official sourceSource: ChatGPT yayın notlarıCyber Verification Program expanded
Anthropic launched an expanded Cyber Verification Program, offering verified security professionals more advanced cyber capabilities and less restrictive classifiers. The program aims to enable more controlled use of new models in defensive security work.
Usage details
Who can use it: Cybersecurity professionals and organizations whose eligibility is verified; an application and program approval are required.
How to use it: Verify your organization and intended use through the Cyber Verification Program application process.
Tip: Clearly document the defensive scope, human approval points, and abuse-prevention controls in your application.
✓ Verified by official sourceModels: Claude Opus 5.5, Claude Sonnet 5.5, Claude Mythos 5.1Source: Cyber Verification Program genişlemesiGrok 4.7 generally available in Microsoft Foundry
Grok 4.7 is generally available in the Microsoft Foundry Models catalog. Developers can connect the model to enterprise applications and agents using Chat Completions or Responses API, tool calling, and streaming.
Usage details
Who can use it: For Microsoft Foundry users in Global Standard and US Data Zone Standard deployments; short-context pricing starts at 2 US dollars for input and 6 US dollars for output per 1 million tokens.
How to use it: Select and deploy the `grok-4.7` model in the Foundry Models catalog; call it in your application using Foundry APIs and Microsoft Entra ID or an API key.
Tip: Teams with US Data Zone requirements should also test long-context and regional pricing differences when choosing a deployment region.
Nano Banana 2.1 became generally available
Nano Banana 2.1 reached GA with improved image quality, prompt adherence, multi-turn character consistency, and text generation. The model supports 1:4, 4:1, 1:8, and 8:1 panoramic aspect ratios, as well as 1K, 2K, and 4K resolutions.
Usage details
Who can use it: Rolling out across Gemini app, Google AI Studio, Gemini API, Google Search AI Mode, Google Ads, Google Flow, and Google Stitch; regional and plan restrictions vary by product.
How to use it: Select the gemini-nano-banana-2.1 model in the API; in the Gemini app, prompt it to generate or edit an image.
Tip: For character or product consistency, explicitly specify reference images and the target aspect ratio in the prompt.
✓ Verified by official sourceModels: Gemini Nano Banana 2.1, gemini-nano-banana-2.1Source: Release notes | Gemini API - Google AI for DevelopersSource: Nano Banana 2.1API usage tiers simplified from five to three
OpenAI has reduced its API usage tiers from five levels to three: Build, Launch, and Grow. Organizations are automatically upgraded when their total credit purchases reach the relevant thresholds; monthly limits and rate limits per model can be tracked on the new tier page.
Usage details
Who can use it: All OpenAI API organizations; automatic tier upgrades depend on credit purchases.
How to use it: In the API Platform, check whether your organization is at the Build, Launch, or Grow level on the Usage tiers or rate limits page.
Tip: Before buying credits unnecessarily to move up a tier, first measure your actual monthly usage and rate limit needs.
✓ Verified by official sourceSource: OpenAI API changelogFinances expands to Free and Go users
ChatGPT Finances is beginning to roll out to Free and Go users in the U.S. Users can securely connect their financial accounts to review spending, savings, and investments with answers based on their own data.
Usage details
Who can use it: Free and Go plans on web, iOS, and Android in the U.S.
How to use it: Open the Finances experience in ChatGPT and connect supported financial accounts.
Tip: Before making financial decisions, check which connected accounts are covered and when the data was last updated.
✓ Verified by official sourceSource: ChatGPT yayın notlarıGemini 3.1 Flash Image entered the deprecation process
The Gemini 3.1 Flash Image model was deprecated, and Google asked users to switch to Nano Banana 2.1 for new image generation and editing tasks. No shutdown date has been announced yet.
Usage details
Who can use it: Gemini API developers; a shutdown date for the old model has not yet been announced.
How to use it: Use the gemini-nano-banana-2.1 model ID instead of gemini-3.1-flash-image in requests.
Tip: Migrate after testing image quality, resolution, and cost differences with your production use cases.
✓ Verified by official sourceModels: gemini-3.1-flash-image, gemini-nano-banana-2.1Source: Release notes | Gemini API - Google AI for DevelopersSource: Gemini deprecationsModels API reports server_tools capabilities
The capabilities.server_tools field was added to the Models API. Developers can now dynamically check from the model list whether a model accepts web search or code execution tools.
Usage details
Who can use it: Claude Models API users.
How to use it: Read the capabilities.server_tools.web_search and code_execution fields in the response to GET /v1/models or GET /v1/models/{model_id}.
Tip: Automatically configure the toolset based on these fields instead of including unsupported tools in the request body.
✓ Verified by official sourceSource: Claude Platform API sürüm notlarıIn-product BAA flow for HIPAA compliance in the API
Eligible organization administrators can now accept the standard Business Associate Agreement and enable HIPAA compliance support from API Organization settings > General. This brings the contract and configuration process for health data workflows into the Platform.
Usage details
Who can use it: Eligible API organizations; scope and service eligibility depend on OpenAI’s BAA terms.
How to use it: In the API Platform, start the BAA acceptance and HIPAA support flow from Organization settings > General.
Tip: Before enabling compliance, confirm with your legal and security teams which endpoints and data flows are covered.
✓ Verified by official sourceSource: OpenAI API changelogFive new interface languages added to Grok Bot
Marathi, Tamil, Telugu, Swahili, and Hausa were added to Grok Bot settings. This usability expansion makes not only the Bot’s response language but also its app interface available to more users.
Usage details
Who can use it: In Grok Bot’s Language options, for supported accounts and client versions.
How to use it: Select Marathi, Tamil, Telugu, Swahili, or Hausa from Settings → Language.
Tip: If you use Follow System to match the system language, make sure your computer’s language is one of the supported options.
✓ Verified by official sourceSource: Grok Bot ChangelogGrok 4.7 powers Expert and Fast modes
Grok 4.7 is now the model behind Expert mode on grok.com and Fast mode for paid subscribers. This lets users choose between more powerful coding and knowledge work and faster responses in the web app.
Usage details
Who can use it: Expert mode is available on grok.com; Fast mode is available to paid subscribers.
How to use it: In a grok.com chat, select Expert or, if you’re on a paid plan, Fast from the model/mode selector.
Tip: Try Expert for lengthy coding and research tasks, and Fast for short, time-sensitive questions.
Grok Bot expands to SuperGrok Plus on grok.com
Grok Bot has added SuperGrok Plus subscribers to the experience previously available to SuperGrok Heavy subscribers on grok.com. Bots can now be opened from the Bots section of the web app’s sidebar and continued in the browser.
Usage details
Who can use it: SuperGrok Plus and SuperGrok Heavy subscribers.
How to use it: Sign in to grok.com; open an existing Bot or chat thread from the Bots section of the sidebar.
Tip: Enable browser notifications so you don’t miss a Bot’s completed tasks while the tab is in the background.
✓ Verified by official sourceSource: Grok Release NotesGrok Voice Transcribe 1.0 discontinued
`grok-voice-transcribe-1.0` was deprecated and reached end of life on October 2, 2026. Requests to the old slug are being routed to the `grok-voice-transcribe-2.0` model, which offers higher accuracy at the same price.
Usage details
Who can use it: Affects API customers using the old 1.0 slug; routing is applied automatically at the same price.
How to use it: Set `model` to `grok-voice-transcribe-2.0` in your client code, or use the existing default routing.
Tip: After automatic routing, compare the language, timestamps, and diarization output against your production samples again.
✓ Verified by official sourceModels: grok-voice-transcribe-1.0, grok-voice-transcribe-2.0Source: Release NotesUnified editing with up to 10 references in Qwen-Image-2.1-Pro
Qwen-Image-2.1-Pro combines text-to-image generation and editing in a single model with a 7B image-generation component. Key new features include transparent image generation, up to 10 reference images, local mask editing, and consistency for people and products.
Usage details
Who can use it: Available through the Model Studio International API.
How to use it: Select the qwen-image-2.1-pro model in a Model Studio request; include reference images and local mask inputs in the request.
Tip: For an e-commerce image, provide the product, model, and accessories as separate references and combine them in a single composition.
✓ Verified by official sourceModels: qwen-image-2.1-proSource: Qwen-Image-2.1: Compact, Efficient, and Unified Image CreationSource: Model lifecycle and updatesVirtual try-on and favorite folders arrive in ChatGPT Shopping
For clothing and accessories, users can tap the Try on button and use a selfie for a virtual try-on. Users can also save products to Favorites and organize them into folders; the reference photo can be managed in Personalization settings.
Usage details
Who can use it: In ChatGPT’s web and mobile apps.
How to use it: Tap Try on in a product listing and add a photo. Use Favorites or folders to save products.
Tip: For more consistent comparisons, use a reference photo taken from the front in the same lighting as your virtual try-on.
Marketplace renamed Connect Apps
Marketplace in Grok Bot has been renamed Connect Apps, highlighting category chips, personalized recommendations, and Bot templates on the web. Connector sign-in and app permissions are also managed through clearer cards.
Usage details
Who can use it: Grok Bot desktop and web users; the apps shown may vary based on account and team permissions.
How to use it: Open Connect Apps in Grok Bot, select an app by category or search, and complete the connection.
Tip: Before connecting a connector, check the card to see which account and permissions the Bot will use.
✓ Verified by official sourceSource: Grok Bot ChangelogMulti-page Scan mode for the mobile ChatGPT camera
Scan mode in the mobile camera captures multiple pages in succession and combines them into a single PDF. This makes handwritten notes and documents ready to upload directly into a chat.
Usage details
Who can use it: Rolling out on iOS.
How to use it: In mobile ChatGPT, open the camera, select Scan from the three-dot menu, and capture the pages in succession.
Tip: Fit the document edges fully in the frame and capture pages in a consistent order; check the preview before the PDF is created.
✓ Verified by official sourceSource: ChatGPT yayın notlarıPrimary Bot offers proactive task suggestions
The starred Primary Bot can spot tasks it could take off the user’s plate and offer to handle them. This moves the Bot beyond being an assistant that waits for prompts toward proactively tracking tasks.
Usage details
Who can use it: Users with access to Grok Bot; Primary Bot setup required.
How to use it: Use the starred Primary Bot during Grok Bot setup or in Bot details; approve the task from the suggested task card.
Tip: Start with low-risk, repetitive tasks for proactive suggestions; require approval for external communications and financial decisions.
✓ Verified by official sourceSource: Grok Bot ChangelogInterface and behavior extensions with Claude Code Mods
Claude Code plugins can now change deeper behaviors using JavaScript or TypeScript event handlers. Mods can render custom panels and commands, pause or transform tool calls, and run across Claude Code’s CLI, Desktop Code tab, and VS Code surfaces.
Usage details
Who can use it: Claude Code 2.1.287 and later; trusted marketplace plugins and local plugins.
How to use it: Install a mod plugin or install one from the marketplace with /plugin install; run /reload-plugins when needed.
Tip: Because Mods run code with your permissions, install them only from sources you trust and review the manifest.
✓ Verified by official sourceSource: Claude Code 2.1.287 sürüm notlarıSource: Claude Code Mods dokümantasyonuModel line field added to Models API
GET /v1/models and model detail endpoints now return each model’s family in the line field. Developers can group families such as Opus without parsing model IDs.
Usage details
Who can use it: Claude Models API users.
How to use it: Read the line field in the model list response; its value may be null for models that do not belong to a family.
Tip: For routing that can handle new model releases, use the line field instead of extracting family names from ID strings with regex.
✓ Verified by official sourceSource: Claude Platform API sürüm notlarıRay-Ban Meta (Gen 3) launched in India
Ray-Ban Meta (Gen 3) launched in India with a slimmer design, up to nine hours of battery life, a customizable action button, a 12 MP camera, and 3K Ultra HD video. Meta also announced Ray-Ban Meta Audio and Muse support for AI glasses in the coming months.
Usage details
Who can use it: Available in India at meta.com, ray-ban.in, and select electronics and optical stores; starting price 44.300 INR. Ray-Ban Meta Audio coming soon.
How to use it: Buy Gen 3 from Meta or Ray-Ban stores in India and pair it with the Meta AI app.
Tip: Configure noise reduction from the six microphones for crowded or windy environments, and set up the action button for quick Meta AI access.
✓ Verified by official sourceModels: Ray-Ban Meta (Gen 3)Source: Announcing Ranveer Singh as Brand Ambassador for Ray-Ban and Ray-Ban Meta in India along with Exciting New Updates to our AI GlassesDreams research preview adds support for new models
The Dreams research preview in Managed Agents now supports Claude Opus 5.5, Claude Fable 5.1, and Claude Sonnet 5.5. These models can now be used in the Dreaming workflow, which reviews past agent sessions and generates suggestions for new workflows.
Usage details
Who can use it: Dreams research preview; Managed Agents access and an account for a supported model are required.
How to use it: Select a model from a supported model family in the Dreams configuration.
Tip: Have a person review automatically generated workflow suggestions before adding them to production Skills.
✓ Verified by official sourceModels: Claude Opus 5.5, Claude Fable 5.1, Claude Sonnet 5.5Source: Claude Platform API sürüm notları
September 2026
Claude Sonnet 4.5 announced for deprecation
Anthropic announced that the claude-sonnet-4-5-20250929 model will be deprecated and set November 30, 2026, as the Claude API retirement date. Developers were advised to move to Claude Sonnet 5.5.
Usage details
Who can use it: The model will be available in the Claude API until November 30, 2026; it is planned to be retired from the API afterward.
How to use it: Replace the model ID with claude-sonnet-5-5 and apply the differences in thinking and tool_choice documented in the Sonnet 5.5 migration guide.
Tip: Before the retirement date, compare staging and production responses using the same evaluation set.
✓ Verified by official sourceModels: Claude Sonnet 4.5, Claude Sonnet 5.5Source: Claude Platform API sürüm notlarıGemini 4 Argon first opened to trusted cyber defenders
Google announced Gemini 4 Argon, with a 1 million token output limit and capabilities for long-running coding, enterprise knowledge work, and cyber defense. To strengthen security measures in stages, initial access was limited to trusted cyber defenders under the Fairwind Program.
Usage details
Who can use it: Trusted cyber defenders and testers under the Fairwind Program; developer, business, and consumer access will follow in later stages.
How to use it: Until general access opens, eligible users can apply through the Fairwind Program; general users should follow the app/API announcement on October 13.
Tip: While model access is limited, document security assessments and usage limits before incorporating it into real workflows.
✓ Verified by official sourceModels: Gemini 4 ArgonSource: Gemini 4 Argon: our next era of frontier intelligenceCollaborative human–AI editing added for Notion and Feishu documents
Kimi Work 3.2.15 added the ability to open Notion and Feishu documents from the homepage sidebar, display documents created with plugins as artifact cards, and support comments and page Q&A in the built-in browser.
Usage details
Who can use it: Kimi Work users who use Notion or Feishu; the relevant plugin and authorization are required.
How to use it: Open a Notion or Feishu document from the homepage sidebar; if a plugin created it, select its artifact card.
Tip: Check document comments and page context before AI editing to avoid changes being made in the wrong section.
✓ Verified by official sourceSource: Kimi Work Release NotesSkills introduced to replace Gems
Skills were introduced as reusable custom instructions in the Gemini app and most Workspace apps; multiple Skills can be layered in the same prompt. Google announced that it will phase out Gems support and convert remaining Gems into draft Skills in the Gemini app.
Usage details
Who can use it: Personal and Workspace accounts in the Gemini app; eligible Business, Enterprise, and education licenses for Workspace Skills, with an 18+ requirement.
How to use it: Use the Skills creation help flow in the Gemini app; in Workspace, create a separate Skill through Skill Builder.
Tip: For recurring tasks, save brand voice, format, and quality criteria in a Skill, then layer different Skills in the same conversation.
Admin API organization methods exit beta
Admin API methods for organization information, members, invitations, workspaces, API keys, rate limits, service accounts, workload identity federation, customer-managed keys, and compliance settings have exited beta.
Usage details
Who can use it: Claude Platform developers; stable client.organization and ant organization surfaces are available in new SDK releases and the ant CLI.
How to use it: Upgrade to the relevant version of the Python, TypeScript, C#, Go, Java, PHP, or Ruby SDK and use client.organization.
Tip: Rather than removing old beta calls immediately, compare response schemas with the stable methods for a while.
✓ Verified by official sourceSource: Claude Platform API sürüm notlarıTeam Bot managers and avatar customization
The creator of a published Team Bot can assign teammates as Managers to jointly manage profile, skill, plugin, secret, and Slack app settings. Bot avatars can also be customized with new shape options and a refreshed look.
Usage details
Who can use it: Team Bot owners and authorized managers on Teams and Enterprise plans.
How to use it: Add a teammate from the Managers section in Team Bot details; choose a shape and color in the avatar editor.
Tip: Grant the manager role only to people who will oversee the Bot’s permissions and connector scope.
✓ Verified by official sourceSource: Grok Bot ChangelogVoice chats with Team Bots and automatic updates enabled by default
Teammates can now start voice chats with a Team Bot in their own private conversations. Automatic Updates is enabled by default in the Grok Bot desktop app, allowing updates to be applied while the user is away.
Usage details
Who can use it: Teams and Enterprise plans using Team Bots; automatic updates are available in the desktop app.
How to use it: Use the voice call button in a Team Bot chat; you can turn off automatic updates in Settings → Updates.
Tip: If update timing is critical on corporate computers, align the Automatic Updates setting with your IT policy.
✓ Verified by official sourceSource: Grok Bot ChangelogComputer use added to the Agents API
The Agents API can now perform tasks in an OpenAI-hosted browser. The application manages access approvals and sign-in flows as agents interact with websites; the feature brings Codex’s computer-use capabilities to the API.
Usage details
Who can use it: Developers using the Agents API; also available in Codex and ChatGPT Work on Pro 500 and Enterprise plans.
How to use it: Enable the computer-use tool in the Agents API and manage site access approvals and authentication flows in your application.
Tip: Give the browser agent access only to the domains it needs, and require explicit user approval before irreversible actions.
$500 Pro 500 plan with Astra Ultrafast access
OpenAI has introduced the $500-per-month Pro 500 plan. It includes the highest usage allowance among the Pro tiers and access to GPT-6 Astra Ultrafast in ChatGPT Work and Codex; the current plan page also lists GPT-6.1 Sol Ultrafast.
Usage details
Who can use it: New Pro 500 subscriptions are subject to account eligibility; Ultrafast is available from the plan selector.
How to use it: In ChatGPT, select Pro 500 under Settings > My Plan; turn on Ultrafast from the model selector in Work or Codex.
Tip: Since Ultrafast uses its allowance faster than standard mode, use it only for latency-sensitive tasks.
✓ Verified by official sourceModels: GPT-6 Astra, GPT-6.1 SolSource: ChatGPT Pro kademeleriSource: ChatGPT yayın notlarıCodex Cloud, refreshed CLI, Code Review, and Security Cloud
At DevDay, OpenAI announced Codex working with reusable environments in the cloud; a refreshed CLI with voice guidance and an `/agents` view; Code Review for inspecting GitHub/GitLab changes on desktop; and Codex Security Cloud for scanning repositories and preparing fixes in the cloud. Together, these expand coding and security work that can continue away from local development.
Usage details
Who can use it: Codex Cloud on Plus, Pro, Business, Healthcare, Education, and Enterprise plans; the new CLI on all plans; Security Cloud on Pro, Business, Enterprise, and Edu.
How to use it: Set up a repository and reusable environment in Codex Cloud; open the `/agents` view in the CLI, Code Review on desktop, or Security Cloud on the web or desktop.
Tip: Pin approved tools, dependencies, and permissions in reusable environment templates for cloud tasks.
Dots introduced as always-on agents
Dots were announced as agents that learn a user’s priorities and can take on ongoing responsibilities by working continuously. The product is positioned as a new surface where people and agents work together.
Usage details
Who can use it: Pro and Business Premium in eligible markets; beta in Enterprise, Edu, and Healthcare workspaces with admin enablement.
How to use it: Create a dot in the ChatGPT desktop app or on the web, connect your apps, and assign it a task from its profile.
Tip: Before giving a dot ongoing access, limit its task scope, connected apps, and which actions require approval.
✓ Verified by official sourceSource: DevDay 2026 özetiPlugin Extensions, the builder, Sites, and MCP Events expanded
OpenAI has launched Plugin Extensions, allowing developers to create interactive panels and file viewers in the ChatGPT sidebar. The plugin creation, submission, and discovery flow has been updated; Sites can host plugins, and MCP Events can trigger automations based on events in connected apps.
Usage details
Who can use it: Plugin Extensions, creation, and MCP Events are available on all plans; plugin use in Sites is available on Business, Enterprise, Healthcare, and Edu.
How to use it: Developers can set up Extensions and MCP Events flows using the Plugins developer documentation; Site owners can add supported plugins to a Site.
Tip: Publish plugin actions with only the permissions they need, and limit automation triggers to reversible actions.
✓ Verified by official sourceSource: DevDay 2026 özetiAPI Ultrafast service tier for GPT-6 Astra
Ultrafast mode is now available for GPT-6 Astra in the Responses API. `service_tier: ultrafast` offers lower token latency than standard processing; global processing and US data residency are supported, while EU and other regional inference residency options are not.
Usage details
Who can use it: API customers; subject to rate limits, EU and other regional inference residency are unavailable.
How to use it: Send `gpt-6-astra` with `service_tier: "ultrafast"` in a Responses request.
Tip: For workloads that require EU data residency, use Standard or another supported model.
✓ Verified by official sourceModels: GPT-6 Astra, gpt-6-astraSource: OpenAI API changelogSource: DevDay 2026 özetiGPT-6.1 Sol offers capabilities close to Astra at lower cost
GPT-6.1 Sol was released as an upgrade that comes close to GPT-6 Astra in agentic coding, computer use, and professional work. Its standard API pricing is $2 per 1 million input tokens, $0.10 per 1 million cached input tokens, and $10 per 1 million output tokens; the model also includes beta multi-agent support.
Usage details
Who can use it: Plus, Pro, Business, Enterprise, and Edu in ChatGPT Work and Codex; also available in the OpenAI API as gpt-6.1-sol.
How to use it: Select GPT-6.1 Sol from the model list in Work or Codex; use `gpt-6.1-sol` in the API.
Tip: For long coding tasks, measure cached context costs alongside lower reasoning effort levels.
✓ Verified by official sourceModels: GPT-6.1 Sol, gpt-6.1-solSource: GPT-6.1 Sol duyurusuSource: OpenAI API changelogGPT-Live daily limits simplified and separate intelligence levels removed
New daily limits were announced: three hours with GPT-Live-1 mini for Go, three hours with GPT-Live-1 for Plus, 15 hours for Pro 100, and unlimited use for Pro 200. Automatic switching to the mini model after reaching the limit was removed for Plus and Pro; the Instant, Medium, and High Voice intelligence levels were also retired.
Usage details
Who can use it: On Go, Plus, and Pro plans; actual limits depend on the account, abuse safeguards, and current plan terms.
How to use it: Open ChatGPT Voice as usual; use your plan’s GPT-Live limit instead of selecting a separate intelligence level.
Tip: Track limit usage during long conversations, and have a plan to switch to text-based ChatGPT if the task is critical.
Library file and folder sharing
Library sharing lets users share files and folders in ChatGPT with specific people or an entire workspace. Viewer and Editor permissions, access management, and the Shared with me area make it possible for teams to chat about shared content.
Usage details
Who can use it: Sharing features depend on the account, plan, and workspace permissions.
How to use it: Open the sharing dialog for a file or folder in Library, then select access for a person or workspace and the Viewer/Editor role.
Tip: For shared folders, check in advance who will retain the file if ownership changes or access is removed.
✓ Verified by official sourceSource: ChatGPT yayın notlarıMuse for Small Business and business connectors launched
Meta launched the Muse for Small Business experience, which works in the background when given a goal by a small business. Instagram professional account analytics, Facebook Pages, and ad accounts can be connected in a few steps, along with tools such as Asana, Box, Canva, Dropbox, Figma, QuickBooks, Klaviyo, Notion, Shopify, Slack, Stripe, and Zoom.
Usage details
Who can use it: In the US and Canada, where Meta offers Muse; basic use is free, with subscription plans available for higher usage. Connectors may vary by account and region.
How to use it: In the Muse app, connect business tools and Meta business accounts from the connectors list in settings, then provide a business goal.
Tip: For the first task, ask it to create a measurable growth plan for the coming period using sales, advertising, and social media data.
✓ Verified by official sourceModels: Muse SparkSource: The Future Is for Everyone: Muse for Small BusinessApproval cards and private Bot-to-Bot routing in Slack
A Review an action card has been added to Slack threads for actions requiring approval; only an authorized person can respond. Teams can also ask a Team Bot in Slack to message their own Bot and receive the result privately in Grok Bot.
Usage details
Who can use it: Teams using Team Bots and a Slack connection; admin approval and channel permissions are required.
How to use it: Add the Team Bot to a Slack channel; respond to the card when approval is required, or ask the Team Bot to send a message to another Bot.
Tip: Grant approval authority only to people who will assess the action; in public channels, prefer sending sensitive results to a private message.
✓ Verified by official sourceSource: Grok Bot Changelog@ChatGPT in Slack/Teams, the Meetings add-on, and shareable profiles
@ChatGPT is now available in Slack and Microsoft Teams for use in channels, threads, or DMs. The Meetings add-on generates notes and action summaries on desktop; shareable profiles bring Sites and apps together in one discoverable place.
Usage details
Who can use it: @ChatGPT for Business and Enterprise; Meetings beta on macOS desktop for Pro and Business; profiles for Free, Go, Plus, Pro, Business, and Enterprise.
How to use it: Ask your admin to enable @ChatGPT; open Meetings in the macOS desktop app and profiles in your ChatGPT profile area.
Tip: Align retention and sharing settings for sensitive audio data in meeting notes with your team policy.
A collaborative workspace with Space, Pages, and team tasks
ChatGPT Space was announced as a new space where teams can work with shared information in ChatGPT and dot; Pages supports collaborative editing, research, and graphic and image generation. Teams can also schedule recurring tasks or trigger them with events such as emails and Slack.
Usage details
Who can use it: Space and Pages are available on Pro, Business, and Enterprise plans; team tasks are available on Business and Enterprise plans.
How to use it: Create a Space or Page in ChatGPT and invite team members; configure team tasks with connected tools and schedules.
Tip: Check Space, Page, and task permissions separately before sharing personal chats in a shared workspace.
OpenAI Marketplace for Enterprise commitments
Eligible Enterprise customers can apply to allocate part of their existing OpenAI commitment to approved partner software through the Marketplace. Initial partners include Figma, Adobe, Salesforce, ServiceNow, Harvey, Legora, Palo Alto Networks, and CrowdStrike.
Usage details
Who can use it: Eligible Enterprise customers; subject to application and approval.
How to use it: Submit an enterprise application through the OpenAI Marketplace interest form.
Tip: Before purchasing, clarify the partner software’s data access, license scope, and terms for applying charges against the OpenAI commitment.
✓ Verified by official sourceSource: DevDay 2026 özetiMuse expands to Canada as well as the US
Meta announced that Muse's personal agent experience is available in the US and Canada. This access was confirmed alongside the Muse for Small Business announcement and also makes personal tasks and business connectors available in these markets.
Usage details
Who can use it: US and Canada; iOS, Android, web, and surfaces supported by Muse. Basic use is free, with subscription plans available for higher usage.
How to use it: Open the Muse app or muse.ai and check availability in your account region.
Tip: For business use in Canada, first verify the Instagram professional account, Facebook Page, and email calendar connections separately.
✓ Verified by official sourceModels: Muse SparkSource: The Future Is for Everyone: Muse for Small BusinessClaude Sonnet 5.5 released
Claude Sonnet 5.5, the second Claude 5.5 model, is more than 30 percent faster than Sonnet 5 and costs up to 30 percent less per task for most work. It handles coding, documents, presentations, spreadsheets, and visual design tasks in fewer turns.
Usage details
Who can use it: Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud, and Microsoft Azure; Türkiye is on the list of supported regions, with no dedicated Türkiye rollout specified. Text watermarking is applied to new models in the EU.
How to use it: Use the claude-sonnet-5-5 model ID on Claude Platform; in Claude apps, select Sonnet 5.5 from the model picker.
Tip: Start with Low or Medium effort for routine coding and document tasks, and increase effort only for complex tasks.
✓ Verified by official sourceModels: Claude Sonnet 5.5Source: Claude Sonnet 5.5 duyurusuSource: Claude Platform API sürüm notlarıSource: Desteklenen bölgeler politikasıTeam Bots enters public beta
Team Bots launched as shared AI coworkers that use the same files, apps, skills, and Bot memory across a team. Each team member’s conversations remain private, while the Bot can use shared context for the team and, when needed, the Slack workspace.
Usage details
Who can use it: Public beta on Teams and Enterprise plans.
How to use it: Configure a Grok Bot for your team, add the required plugins, credentials, files, and skills, then share it as a Team Bot.
Tip: Give the Team Bot only the files and permissions it needs for its role; test the sharing settings to confirm that personal conversations remain private.
✓ Verified by official sourceSource: Team Bots: shared AI teammates that learn as they workSource: Grok Bot ChangelogMeta Enterprise Platform launched
Meta launched Meta Enterprise Platform to productize its AI stack for businesses and developers. The initial scope aims to turn the Muse agent, Meta Business Agent, Muse API, and Muse Code into services that businesses can use in their own operations.
Usage details
Who can use it: A new platform for businesses and developers; detailed plan, regional, and pricing information has not yet been provided in the announcement.
How to use it: Follow Meta Enterprise Platform channels for enterprise access and product announcements; use Meta Model API and Muse Code for current developer access.
Tip: When planning enterprise use, document model access, data privacy, and agent approval workflows as separate requirements.
Qwen App’s Quark Cloud Drive integration reported
According to AIbase, after user authorization, Qwen App connected Quark Cloud Drive files to in-chat search, editing, and reading workflows. The report also describes generating study materials, documents, or interactive pages from files in the same workflow and creating shareable links; however, no independent official changelog page could be found for this entry.
Usage details
Who can use it: Reported only for the China ecosystem workflows mentioned in the article, involving Quark and Qwen; access in Türkiye/EU has not been confirmed.
How to use it: The news source describes invoking Quark Cloud Drive with @ on the Qwen home page and, if necessary, enabling the Quark Cloud Drive Management skill.
Tip: Before granting access to cloud files, check which folders Qwen will be able to read and write.
Loop video generation in Imagine
The Loop tool added to the Imagine home page generates a video that starts and ends with the selected image. The feature can also be opened from an image’s Make Video menu, providing more control over transition animations.
Usage details
Who can use it: Grok Imagine users; the plan’s video-generation allowance and rate limits apply.
How to use it: Select the Loop tile on the Imagine home page, or open an image’s Make Video menu and use Loop.
Tip: For scenes suited to looping, such as product rotations or location transitions, keep the start and end images in the same composition.
Editable email drafts in Outlook
Grok Bot can now edit new email drafts it creates in an Outlook account within the chat and send them after user approval. This brings corporate email workflows outside Gmail into the same draft-and-approval experience.
Usage details
Who can use it: Grok Bot users; a connected Outlook account and sending permissions are required.
How to use it: Connect Outlook through Connect Apps, ask the Bot to prepare a draft, and use the Send email option on the chat card.
Tip: When using it for the first time, check the sender address, recipients, and signature on the draft card.
✓ Verified by official sourceSource: Grok Bot ChangelogThe `safety_identifier` field in the API
Developers can now send an opaque `safety_identifier` in requests to associate policy violations with an application’s end user rather than its API key. The field is accepted in Chat Completions, Responses, deferred completions, Batch API, and gRPC requests.
Usage details
Who can use it: Available to developers using the API; the legacy `user` field continues to be accepted.
How to use it: Add an opaque value generated by your application for each end user to the `safety_identifier` field.
Tip: Do not put an email address or directly identifying personal data in the identifier field; use an irreversible, opaque application ID.
✓ Verified by official sourceSource: Release NotesClaude plugin submission portal opens
Developers can now submit plugins containing an MCP connector or Skills through the new directory portal, track their security scan and review status, and view installation and discovery analytics after publication. Plugins are listed in Claude Directory.
Usage details
Who can use it: Developers on paid Claude plans; a single MCP connector or a plugin bundle hosted on GitHub can be submitted.
How to use it: Submit your plugin at claude.ai/directory/manage/new, fix the automated validation results, and publish it after approval.
Tip: Run the pre-submission test suite to validate MCP 2.0, MCP Apps, and Enterprise Managed Auth extensions early.
✓ Verified by official sourceSource: Claude için eklenti oluşturmaSource: Claude uygulama sürüm notlarıImage encoding bug fixed in GPT-6 Sol and Luna
OpenAI fixed an image encoding bug that impaired image understanding in GPT-6 Sol and GPT-6 Luna. The fix aims to improve results on image tasks and computer-use workflows in the API and Codex.
Usage details
Who can use it: API and Codex users; the new model behavior is applied automatically.
How to use it: Rerun workflows that use image inputs; repeat your evaluations if needed.
Tip: For image evaluations, compare old and new outputs using the same test set.
Interrupted Qwen Code workflows resume where they left off
Long-running workflows now save the status of each step. When reopened, completed steps are preserved without being rerun; only the failed path can be retried, or the entire workflow can be restarted if desired.
Usage details
Who can use it: Qwen Code v0.24.1–v0.24.4; Web Shell workflow users.
How to use it: In Web Shell, go to Workflows history and select either Retry Failed Path or Rerun All.
Tip: For workflows that take hours, retry failed subagents first instead of rerunning everything.
✓ Verified by official sourceSource: Qwen Code Weekly: See Which Files Fill Your Context, Edit Projects over SSH, Resume Interrupted WorkflowsMeta AI Connectors developer preview opens
Meta AI Connectors was announced as an integration layer that lets developers connect their services to Meta AI chat through an API or MCP server. Users will be able to use the same connection across Meta AI surfaces on mobile, web, and AI glasses without switching apps.
Usage details
Who can use it: Developer preview; early access registration is required, and access is rolling out gradually.
How to use it: Sign up at dev.meta.ai/products/connectors and define the actions on your API or MCP server that Meta AI can call.
Tip: Enable only the actions needed in the connector, and require explicit user approval for write operations.
✓ Verified by official sourceSource: Meta Connect Recap: How To Build For AI Glasses And Reach an AudienceMeta Model API reaches global general availability
Meta Model API became available globally with general access following Connect 2026. Meta said it offers support and compliance for enterprise workloads through its first-party API and third-party partners, that Muse Spark 1.3 is coming to Oracle Cloud AI Platform, and that it is available in private preview on Google Cloud.
Usage details
Who can use it: Global developer access; Geographic Use Policy restrictions by country and enterprise compliance requirements apply.
How to use it: Create an API key through dev.meta.ai and connect to Meta Model API endpoints; select an available model ID from the model catalog.
Tip: Check country restrictions and Standard/Contributor data-use terms before moving to production.
✓ Verified by official sourceModels: Muse Spark 1.3, Muse Image, Muse Voice Transcribe, Muse GlimmerSource: Meta Connect 2026: The end-to-end recapMuse Code exits beta and comes to Windows
Muse Code became a generally available product out of beta for long-horizon engineering tasks and is now also available on Windows. By default, its multi-agent setup has parallel workers write and test code in isolated workspaces.
Usage details
Who can use it: Muse Code users; Windows support announced with Connect, alongside macOS and Linux. Plan and usage limits may vary by subscription or usage.
How to use it: Install Muse Code from dev.meta.ai; open a repository in the Windows, macOS, or Linux client and assign a multi-step engineering task.
Tip: Break tasks into independently testable subtasks to prevent parallel workers from conflicting.
Version history management added for Office files
Work 3.2.14 added version history management for Office files, making it possible to track changes and revert to previous versions.
Usage details
Who can use it: Kimi Work users.
How to use it: Open version history management for the Office file in the workspace, then view or restore a previous version.
Tip: Create a version checkpoint before a major AI edit.
✓ Verified by official sourceSource: Kimi Work Release NotesQwen Code adds support for the OpenAI Responses protocol
The new openai-responses provider type supports streaming, tool calls, and reasoning summary through OpenAI’s /v1/responses endpoint. This lets Qwen Code use both Chat Completions and Responses-based providers within the same product.
Usage details
Who can use it: Developers configuring custom providers for Qwen Code.
How to use it: Set the provider to openai-responses or add wireApi: responses to the model entry.
Tip: Test Responses compatibility first with a small, read-only tool call.
✓ Verified by official sourceSource: Qwen Code Weekly: See Which Files Fill Your Context, Edit Projects over SSH, Resume Interrupted WorkflowsQwen Code edits remote projects directly over SSH
Add a single address in the format ssh://user@host/absolute/project to Web Shell to read and write files on a remote computer without Qwen Code installed and run commands there. New workspaces are initially read-only, and Trust is required to write.
Usage details
Who can use it: Qwen Code Web Shell; local, server, or team computers with SSH access.
How to use it: When adding a workspace, enter the SSH address as the directory path; Trust the workspace to write files and run commands.
Tip: Before connecting to a production server, prefer a separate verification user with read-only permissions by default.
✓ Verified by official sourceSource: Qwen Code Weekly: See Which Files Fill Your Context, Edit Projects over SSH, Resume Interrupted WorkflowsSora API retired
The Sora web and app experiences had already been shut down on April 26; with the Sora API’s retirement on September 24, the product’s API access has ended. OpenAI explains how users can export their content and how data will be deleted after retirement.
Usage details
Who can use it: The Sora API is unavailable after September 24, 2026; the web and app experiences are also unavailable.
How to use it: Export Sora content through sora.chatgpt.com/sunset during the final export window.
Tip: Prepare a plan to migrate to an alternative video workflow before leaving legacy Sora integrations in production.
File previews and uploads supported from phones
Work 3.2.14 enabled users to preview Office and HTML files on their phones via Remote Control and upload files to Kimi Work from their phones.
Usage details
Who can use it: Kimi Work Remote Control users.
How to use it: Open Remote Control on desktop, then use the file preview or upload button in the phone session.
Tip: When connected on mobile, verify that the file was added to the correct project on desktop before sending large files.
✓ Verified by official sourceSource: Kimi Work Release NotesWearables Device Access Toolkit 1.0 released
After a year in developer preview, Wearables Device Access Toolkit 1.0 reached stable release. With a single integration, existing mobile apps can work on AI glasses with Hey Meta voice invocation, automatic speech recognition, camera capture, audio streaming, motion data, and Meta Neural Band inputs.
Usage details
Who can use it: Available to developers building AI glasses apps; can be used with supported display and displayless glasses.
How to use it: Add the integration to your existing mobile app using the Device Access Toolkit 1.0 documentation in Wearables Developer Center.
Tip: First test workflows that don't require a phone screen with a small prototype that combines camera, audio, and motion data.
✓ Verified by official sourceSource: Meta Connect Recap: How To Build For AI Glasses And Reach an AudienceExplore and discovery surfaces announced for AI glasses experiences
Meta announced new surfaces where selected experiences developed for AI glasses can be discovered on Meta Ray-Ban Display and in the Meta AI app. Developers' connectors and apps will thus move beyond developer mode into a catalog feed that users can find directly.
Usage details
Who can use it: Coming soon; submission and approval details have not yet been released, and selected experiences will be published.
How to use it: Create an account in Wearables Developer Center and submit your experience when approval applications open.
Tip: Design an experience that can be understood with a one-sentence voice discovery prompt on the glasses and completed without looking at a screen.
Side Chat added with inherited session context
Work 3.2.14 added an independent Side Chat to the plus panel that inherits the main conversation’s context.
Usage details
Who can use it: Kimi Work 3.2.14 users.
How to use it: Open the + panel and select Side Chat; continue your question there, independently of the main session.
Tip: Use Side Chat to try an alternative solution or ask a sensitive follow-up question without disrupting the main task.
✓ Verified by official sourceSource: Kimi Work Release NotesDocker/Podman sandbox option for Qwen Code subagents
Subagents can now run in a Docker or Podman container with write access limited to the workspace. The limitations of this backend are clearly stated, including that workflow, Agent Team, MCP, LSP, and skill support are disabled and Windows is not supported.
Usage details
Who can use it: Qwen Code users; macOS/Linux environments with Docker/Podman installed.
How to use it: Set QWEN_AGENT_EXECUTION_BACKEND=docker before launching; verify the sandbox boundaries and disabled features.
Tip: Use this mode when inspecting untrusted repository code; run production workflows that require MCP and skills in a separate session.
✓ Verified by official sourceSource: Qwen Code Weekly: See Which Files Fill Your Context, Edit Projects over SSH, Resume Interrupted WorkflowsQwen Code shows context usage by file and extension
Qwen Code’s /context and /context detail views break down the cost of the system prompt, tools, MCP, memory, skills, and extensions at the file level, including startup context and the cache prefix. Warnings for oversized persistent context and rule loading when a file is touched have also been added.
Usage details
Who can use it: CLI and Web Shell users on Qwen Code v0.24.1–v0.24.4.
How to use it: Run /context or /context detail in a session.
Tip: When context fills up, first reduce the QWEN.md, memory, or extension with the highest token cost.
✓ Verified by official sourceSource: Qwen Code Weekly: See Which Files Fill Your Context, Edit Projects over SSH, Resume Interrupted WorkflowsWebMCP preview for web apps on Ray-Ban Display
Meta added a WebMCP developer preview for web apps running on Meta Ray-Ban Display using standard web APIs. Developers choose which functions Meta AI can call within the app; users can then start those actions through hands-free prompts on the glasses.
Usage details
Who can use it: WebMCP developer preview; for Meta Ray-Ban Display web apps and Meta AI integrations.
How to use it: Expose only the necessary functions in your web app through WebMCP and test them using the Wearables Developer Center documentation.
Tip: Separate functions into read and write operations; require approval for write actions such as purchases, sending, or account changes.
✓ Verified by official sourceSource: Meta Connect Recap: How To Build For AI Glasses And Reach an AudienceWeb page annotation added to the built-in browser
Work 3.2.14 added support for annotating web pages in the browser and moving selected quotes into the input box.
Usage details
Who can use it: Users of the Kimi Work built-in browser.
How to use it: Open a web page in the Work browser, add an annotation, and transfer the quote to the input box.
Tip: In long research sessions, note important passages and send them to the Agent with their source context.
✓ Verified by official sourceSource: Kimi Work Release NotesKeep Awake now runs only on demand
To reduce background power consumption, Work 3.2.14 made Keep Awake opt-in rather than automatic.
Usage details
Who can use it: Kimi Work users.
How to use it: Enable the feature in Power or Keep Awake settings only for tasks that need it.
Tip: Turn it on for long Remote Control sessions; keep it off during everyday use to reduce battery and energy consumption.
✓ Verified by official sourceSource: Kimi Work Release NotesFile watching disabled by default and workspace boundaries tightened
CLI 2.1.0 disabled file watching for config and workspace files by default; the same release introduced stricter security boundaries for symbolic links, untrusted workspaces, and project configuration.
Usage details
Who can use it: Kimi Code CLI users.
How to use it: If hot reload is needed, use [watch] enabled = true or KIMI_CODE_WATCH=1; mark the workspace as trusted.
Tip: Keep watching disabled in untrusted repositories and open only the folder you need as a workspace.
✓ Verified by official sourceSource: Kimi Code CLI ChangelogNew wave of Connected Apps added to Gemini
Gemini began offering new connected apps, including Airtable, Linear, monday.com, PandaDoc, Wispr AI, Zoho, Adobe, Picsart, Squarespace, Webflow, Apartments.com, Experian, Peloton, and SeatGeek. Users can connect them in settings or call them with @ in a chat.
Usage details
Who can use it: Rolling out gradually; each connected app’s country, language, account, and age requirements apply.
How to use it: Connect apps from Gemini Settings or call the relevant app with @ during a chat.
Tip: Before connecting, review what data each app is permitted to read or modify through Gemini.
✓ Verified by official sourceSource: A new wave of Connected Apps is rolling out to Gemini.Muse is coming to AI glasses
Meta announced that Muse will be available hands-free on AI glasses. Users will be able to call Muse by saying their agent's name and ask it to take action on products, listings, or objects in a long list seen by the glasses' cameras.
Usage details
Who can use it: Coming to AI glasses in the coming months; it was not yet generally available at the time of the announcement, and region and model eligibility apply.
How to use it: When Muse support for glasses becomes available for your account, say the agent's name and ask a question or give a task about the object you're looking at.
Tip: Before using information seen by the glasses for actions such as making a purchase or filling out a form, verify the result on screen or by voice.
✓ Verified by official sourceModels: Muse SparkSource: Introducing Ray-Ban Meta Audio and More AI Glasses StylesSource: The Biggest News From Connect 2026Long-form Voice mode announced for Muse
Muse gained a Voice mode that can sustain long, in-depth conversations rather than acting as a voice assistant that responds only to short commands. The agent can work in the background while the user keeps talking; users can also personalize the voice by describing its speed, tone, or accent.
Usage details
Who can use it: Introduced for Muse surfaces in the Connect 2026 announcement; rollout is gradual and regional availability applies.
How to use it: Start a voice conversation in the Muse app and describe your preferred pace, tone, or accent during the conversation.
Tip: Check important steps of a background task with written confirmation before it is completed, rather than during the voice conversation.
Ray-Ban Meta Audio and Gen 3 introduced
Meta opened preorders for its first Ray-Ban Meta Audio glasses, weighing 43 grams and offering up to 12 hours of battery life, at a starting price of 349 US dollars; shipping began on October 13. On the same day, it announced the slimmer Ray-Ban Meta (Gen 3), with up to nine hours of battery life and a customizable action button, at a starting price of 449 US dollars.
Usage details
Who can use it: US preorders for Ray-Ban Meta Audio, with shipments on October 13; Gen 3 on sale on announcement day. Countries and optical options may vary by region; the announcement says expansion to Türkiye will come later.
How to use it: Order through Meta.com or authorized Ray-Ban channels and pair with the Meta AI app.
Tip: Consider the Audio model if battery life and open-ear comfort are priorities, and Gen 3 if the camera and quick AI access are priorities.
✓ Verified by official sourceModels: Ray-Ban Meta Audio, Ray-Ban Meta (Gen 3)Source: Introducing Ray-Ban Meta Audio and More AI Glasses StylesDolby Atmos video capture announced for AI glasses
Meta positioned AI glasses as the first consumer devices to capture video with 360-degree spatial audio using Dolby Atmos. The goal is for users to start recording without adjusting audio settings and play it back later with immersive sound.
Usage details
Who can use it: Coming to supported AI glasses via a software update later in 2026; model and regional availability may vary.
How to use it: When the update reaches your account, start a normal video recording with the glasses’ camera; Dolby Atmos capture will be applied automatically.
Tip: To preserve the spatial audio effect, record in a crowded or active environment and test playback with compatible headphones.
✓ Verified by official sourceSource: Everything We Announced at Meta Connect 2026Hearing enhancement and audio controls coming to AI glasses
Meta announced FDA-approved Hearing enhancement for supported AI glasses to help people with mild to moderate hearing loss. Users will also be able to adjust left-right audio balance or use mono audio mode, which routes all audio to one ear.
Usage details
Who can use it: Coming to the US later in 2026; Hearing enhancement will be available on supported glasses for 149,99 US dollars or with a Meta One subscription, with HSA/FSA eligibility noted.
How to use it: On supported glasses, turn on Hearing enhancement, audio balance, or mono audio in the audio or accessibility settings of the Meta AI app.
Tip: Calibrate hearing enhancement in a quiet environment for your personal comfort, and don’t completely block out surrounding sounds in crowded places for safety.
✓ Verified by official sourceSource: The Biggest News From Connect 2026Experimental full-screen TUI mode added
CLI 2.1.0 introduced an experimental full-screen terminal interface with independent scrolling, collapsible blocks, and a jump-to-bottom button.
Usage details
Who can use it: Kimi Code CLI 2.1.0 users.
How to use it: Select fullscreen in the TUI mode setting in /settings, or use tui_mode = fullscreen in ~/.kimi-code/tui.toml.
Tip: Enable it to navigate long Agent outputs like a document; the standard interface may be faster for short sessions.
✓ Verified by official sourceSource: Kimi Code CLI ChangelogWorkout, nutrition, and shopping AI coming to glasses
Hands-free running and cycling workout guidance, nutrition tracking with personalized insights, and shopping AI that can recognize products, prices, and reviews and provide voice alerts or initiate purchases were announced for Meta AI glasses. These features aim to help users complete everyday tasks without taking their phone out of their hands.
Usage details
Who can use it: Announced at Connect 2026 for release in the coming months; device, language, and regional availability may vary.
How to use it: When the features become available on your account, call up Meta AI on the glasses and give a voice task about running, cycling, nutrition, or a product you’re looking at.
Tip: For shopping, ask for a voice summary comparing prices and reviews, then confirm the purchase separately.
✓ Verified by official sourceSource: Everything We Announced at Meta Connect 2026HTML preview and download management improved
Desktop 1.0.3 opened local HTML files in the transcript as real web pages in the built-in browser; browser downloads gained controls to pause, resume, cancel, and retry.
Usage details
Who can use it: Kimi Code Desktop 1.0.3 users.
How to use it: Click a local HTML file in the transcript; manage the download from the download history in the address bar.
Tip: Verify the web interface created by the Agent in the built-in browser without exporting the file.
✓ Verified by official sourceSource: Kimi Code Desktop ChangelogMeta Ray-Ban Display expands to new European markets
Meta Ray-Ban Display opened to online orders and sales in Canada and the United Kingdom; preorders began in France, Italy, and Germany, with sales starting on October 13. The announcement also said Ray-Ban and Oakley Meta products would become available later in other markets, including Türkiye; no specific date for Türkiye was given on this page.
Usage details
Who can use it: On sale in the United States, Canada, and the United Kingdom; on sale in France, Italy, and Germany starting October 13, 2026. The announcement says “later” for Türkiye.
How to use it: Order through Meta.com or authorized optical/retail channels in supported countries; pair the glasses with the Meta AI app.
Tip: If you plan to buy from Türkiye, wait for stock and compatibility announcements on the official country page; warranty and features may differ for orders from outside the region.
✓ Verified by official sourceSource: New Features for Meta Ray-Ban DisplayMuse Realtime Avatar generates live, expressive avatars
Meta introduced Muse Realtime Avatar, a technology that turns Muse Realtime Voice conversations into real-time avatars with synchronized expressions based on reference images. The system matches face, hand, and full-body movements to the flow of conversation and carries forward generated context to maintain visual consistency throughout long conversations.
Usage details
Who can use it: Introduced as a technology and Muse experience in the Connect announcement; not all example avatars are available in the Muse app.
How to use it: When the avatar feature becomes available for your account in Muse, select a reference image and start a real-time voice conversation.
Tip: Before sharing generated avatars, check Meta's invisible Video Seal watermark and terms of use.
✓ Verified by official sourceModels: Muse Spark, Muse Realtime VoiceSource: Bringing Your Muse to LifeAutomatic session title setting and free-plan warning added
Desktop 1.0.3 added a setting to disable AI naming for new sessions and an upgrade warning when free accounts send a message using a subscription model.
Usage details
Who can use it: Kimi Code Desktop users; the free-account warning appears on applicable plans.
How to use it: Turn off the automatic session title setting in Settings; on the free plan, follow the upgrade prompt when a subscription model is selected.
Tip: Keep title generation off if you don’t want project names changed automatically.
✓ Verified by official sourceSource: Kimi Code Desktop ChangelogAlibaba announces Qwen 4 training and roadmap for future trillion-parameter models
Alibaba announced that Qwen 4 is in training and that future Qwen 4.5 and Qwen 5 series are planned to scale to 5 to 10 trillion parameters. This is a roadmap announcement based on an official company statement about an upcoming model family, not a product launch.
Usage details
Who can use it: Not yet generally available or accessible via API; no release date announced.
How to use it: There is nothing to enable right now; follow Qwen’s official announcements and the Model Studio model catalog.
Tip: Do not treat the roadmap as a replacement for existing Qwen3.8 models; wait before changing production integrations.
✓ Verified by official sourceModels: Qwen 4, Qwen 4.5, Qwen 5Source: Alibaba deepens AI push with new chip, bigger model; shares jump 5%Claude Opus 5.5 released
The first model in the Claude 5.5 family, Claude Opus 5.5, delivers performance approaching Fable 5.1 in agentic coding, computer use, and knowledge work. Compared with Opus 5, it offers about 40 percent lower costs for typical tasks, more than 30 percent higher speed, and lower cache read prices.
Usage details
Who can use it: Claude Platform, Amazon Web Services, Google Cloud, and Microsoft Azure; verification programs apply to high-risk work such as cybersecurity and life sciences.
How to use it: Use the claude-opus-5-5 model ID on Claude Platform; in Claude Code or Claude apps, select Opus 5.5 from the model picker.
Tip: Route context-heavy work such as long, open-ended code migrations to Opus 5.5, and narrow, repetitive tasks to Sonnet or Haiku.
✓ Verified by official sourceModels: Claude Opus 5.5Source: Claude Opus 5.5 duyurusuSource: Claude uygulama sürüm notlarıGemini 3.8 Flash TTS and Flash-Lite TTS reach GA
Google has made two new text-to-speech models generally available. Flash TTS focuses on creative, long-form and multi-turn audio generation, while Flash-Lite TTS focuses on high-volume, low-cost voiceovers and real-time agents.
Usage details
Who can use it: Gemini API and Google AI Studio; rollout has begun in Gemini Notebook and Google Vids, with Gemini Enterprise access coming in a later phase.
How to use it: Select the model in the AI Studio speech generation area or use the model IDs in the Gemini API speech generation guide.
Tip: For long-form content, verify voice and character consistency with a short segment before generating the full piece.
✓ Verified by official sourceModels: Gemini 3.8 Flash TTS, Gemini 3.8 Flash-Lite TTSSource: Release notes | Gemini API - Google AI for DevelopersSource: Gemini 3.8 text-to-speech says helloGPT-6 Sol and Luna come to the API, Work, and Codex
GPT-6 Sol and GPT-6 Luna were released as cost-efficiency-focused sibling models to GPT-6 Astra; they accept text and image inputs and generate text. API prices were announced at $2/$10 and $0.10/$0.50 per 1 million tokens for input/output, respectively; they are available on paid plans in Work and Codex.
Usage details
Who can use it: Plus, Pro, Business, Enterprise, and Edu in ChatGPT Work and Codex; GPT-6 Luna in the desktop app on Free and Go; gpt-6-sol and gpt-6-luna in the API.
How to use it: Select Sol or Luna from the Work or Codex model list; use the relevant model ID in the API.
Tip: Test the same task with Sol and Luna to create task-based routing rules for quality, speed, and token cost.
✓ Verified by official sourceModels: GPT-6 Sol, GPT-6 Luna, gpt-6-sol, gpt-6-lunaSource: GPT-6 Sol ve Luna duyurusuSource: OpenAI API changelogVoices endpoint and custom voice tools released
Support for querying more than 150 preset and custom voices was added to the Gemini API through the /v1beta/voices endpoint. Workflows for designing voices from text, consent-verified voice replication, and creating persistent custom voice personas were also introduced.
Usage details
Who can use it: Gemini API and Google AI Studio; AI Studio voice replication is unavailable in Illinois, Texas, the EEA, the United Kingdom, Switzerland, and India.
How to use it: Query the /v1beta/voices endpoint; follow the consent verification workflow in the voice design or voice replication guide.
Tip: Before replicating a real person’s voice, retain records of explicit consent and usage rights.
✓ Verified by official sourceModels: Gemini 3.8 Flash TTS, Gemini 3.8 Flash-Lite TTSSource: Release notes | Gemini API - Google AI for DevelopersSource: Gemini 3.8 text-to-speech says helloGrok Bot adds support for Czech, Danish, Finnish, Hungarian, and Norwegian
Czech, Danish, Finnish, Hungarian, and Norwegian Bokmål options have been added to Grok Bot’s Language setting. Follow System can also set the app accordingly if the computer uses one of these languages.
Usage details
Who can use it: Available to Grok Bot users through Settings → Language.
How to use it: Select a language in Settings → Language, or turn on Follow System.
Tip: For shared team use, check the client language and the Bot’s response language separately.
✓ Verified by official sourceSource: Grok Bot ChangelogGrok 4.7 released in the API
Grok 4.7 is now available in the xAI API for coding, agent tasks, and knowledge work. The model offers a 500,000-token context window, text and image input, adjustable low/medium/high/xhigh reasoning, and pricing of $2 USD for input and $6 USD for output.
Usage details
Who can use it: For API teams on the global endpoint and US regional endpoint; the applicable long-context price applies to context of up to 500,000 tokens in the public API.
How to use it: Use `model: "grok-4.7"` in a Responses API request; set the `reasoning_effort` field to low, medium, high, or xhigh.
Tip: Reduce cost and latency in long agent loops by using `prompt_cache_key` and, when needed, Context Compaction.
✓ Verified by official sourceModels: grok-4.7Source: Introducing Grok 4.7Source: Release NotesSource: Grok 4.7Grok 4.7 comes to GitHub Copilot
GitHub has begun rolling out xAI’s Grok 4.7 model in Copilot. The model is available in the model picker in VS Code, Visual Studio, Copilot CLI, cloud agent, the Copilot app, JetBrains, Xcode, and Eclipse.
Usage details
Who can use it: Copilot Pro, Pro+, Max, Business, and Enterprise SKUs; phased rollout and Business/Enterprise admin policies apply.
How to use it: Select Grok 4.7 from the model picker in a supported Copilot client; a Business or Enterprise admin can manage access through the Copilot model policy.
Tip: Monitor your team’s quota, taking usage-based provider pricing and the model’s phased rollout into account.
Qwen-Audio-3.1 launches with a four-part ASR suite
Qwen released four Qwen-Audio-3.1 ASR models for short speech, real-time streaming, file transcription, and voice messages. Multilingual and dialect support, hotwords, context boosting, speaker separation, punctuation, and optional text polishing are tailored to different use cases.
Usage details
Who can use it: Model Studio International API users; supported languages and dialects vary by model.
How to use it: Choose the model ID qwen-audio-3.1-asr-flash, -streaming, -filetrans, or -message according to your needs.
Tip: Use filetrans for meeting archives, streaming for live captions, and the message variant for short voice messages.
✓ Verified by official sourceModels: qwen-audio-3.1-asr-flash, qwen-audio-3.1-asr-flash-streaming, qwen-audio-3.1-asr-flash-filetrans, qwen-audio-3.1-asr-flash-messageSource: Model lifecycle and updatesQwen3.8-Omni-Flash-Realtime real-time audio-video API
Qwen3.8-Omni-Flash-Realtime was added as a bidirectional real-time model that generates text and audio responses from streaming audio and video frames. It adds WebSocket, WebRTC, and AOQ access, along with support for function calling, remote MCP tools, web search, multichannel audio, and video aggregation.
Usage details
Who can use it: In Model Studio, Singapore and China (Beijing); a separate API key is required for each region. No separate regional deployment for Türkiye or the EU is specified.
How to use it: For WebSocket, add ?model=qwen3.8-omni-flash-realtime to the workspace URL, or use the WebRTC/AOQ connection flow; configure turn detection with semantic_vad.
Tip: If visual detail is not critical, use video aggregation to reduce real-time inference costs.
✓ Verified by official sourceModels: qwen3.8-omni-flash-realtimeSource: Alibaba Cloud Model Studio: Qwen-Omni-RealtimeSource: Model lifecycle and updatesMore natural two-way conversation with Qwen-Audio-3.1-Realtime-Plus
Qwen-Audio-3.1-Realtime-Plus adds eight new system voices while retaining the previous 3.0 Plus integration protocol. Support for a 262.144-token context, function calling, web search, and voice cloning strengthens real-time voice agents’ ability to complete tasks.
Usage details
Who can use it: API users within Model Studio International.
How to use it: In a real-time connection similar to the 3.0 Plus integration, change the model name to qwen-audio-3.1-realtime-plus.
Tip: If you use voice cloning, verify the reference audio’s authorization and speaker identity in advance.
✓ Verified by official sourceModels: qwen-audio-3.1-realtime-plusSource: Model lifecycle and updatesQwen-Image-2.1 released as open weights
With its 7B image-generation component, Qwen-Image-2.1 combines text-to-image generation and editing in a single model. Key features include transparent backgrounds/layers, up to 10 reference images, mask- or markup-based local editing, and improved consistency for people and products.
Usage details
Who can use it: Open weights are available through Hugging Face, GitHub, and ModelScope; access to the hosted versions on Qwen Chat and Model Studio may vary separately.
How to use it: Use the model weights and sample code in the Qwen-Image-2.1 repository; for the API, select the relevant model ID in Model Studio.
Tip: Combine multiple references into a single composition for a poster or product catalog, then use a mask to limit the final touch-ups.
✓ Verified by official sourceModels: Qwen-Image-2.1Source: Qwen-Image-2.1: Compact, Efficient, and Unified Image CreationSource: Qwen ResearchGrok Bot opens to SuperGrok Heavy on grok.com
SuperGrok Heavy subscribers can now access Grok Bot directly from grok.com. Bots continue running on their own machines, keep working after the user leaves, and multiple Bots can be added to the same group chat.
Usage details
Who can use it: SuperGrok Heavy subscribers.
How to use it: Open the Bots section in the grok.com sidebar; chat directly with a Bot or add multiple Bots to a group.
Tip: Connect the resources the Bot needs without leaving the chat, and enable browser notifications for long-running tasks.
✓ Verified by official sourceSource: Grok Release NotesText Edit, Precise Edit, and video keyframes in Imagine
Grok Imagine added the beta Text Edit feature, which lets users select text in an image and change its wording, font, and color. Precise Edit and multi-region color adjustments are now available on all plans, while paid plans can lock intermediate images as keyframes during video generation.
Usage details
Who can use it: Text Edit beta and Precise Edit are available on all plans; video keyframes are available on paid plans.
How to use it: Open an image in Imagine and select Text Edit or Precise Edit; on a paid plan, lock an image as an intermediate keyframe when editing a video.
Tip: For posters and social media images, use Text Edit to change text instead of regenerating it, preserving character and layout consistency.
Built-in browser began showing live Agent activity and sources
Desktop 1.0.2 brought the Agent’s browser activity into the transcript as summaries with source lists; it highlights the page while a tab is being controlled and can proactively show requested pages.
Usage details
Who can use it: Kimi Code Desktop 1.0.2 users.
How to use it: Open the built-in browser panel and ask the Agent to search for or show a web page.
Tip: For research tasks, make sources visible in the transcript so you can check which pages the Agent used.
✓ Verified by official sourceSource: Kimi Code Desktop ChangelogAgent can use a configured search engine
With Desktop 1.0.2, Agent can search using the search engine configured by the user; the provider settings page was also redesigned.
Usage details
Who can use it: Kimi Code Desktop 1.0.2 users.
How to use it: Configure your search provider in Providers settings, then ask Agent to search in the built-in browser.
Tip: If you need enterprise or regional results, check the search provider before starting the task.
✓ Verified by official sourceSource: Kimi Code Desktop ChangelogBot launches a Project with an on-demand Cloud Agent
When asked to start a Project, a Bot can launch a Cloud Agent that plans the work and runs its own agents. This enables longer-running, parallelizable workflows that go beyond a single chat response.
Usage details
Who can use it: Accounts with access to Grok Bot and the relevant Cloud Agent; plan and team policies apply.
How to use it: Ask the Bot in chat to start a Project and plan the work and distribute it to Cloud Agents.
Tip: Clearly specify the success criteria, file scope, and stopping conditions for the work sent to the Cloud Agent.
✓ Verified by official sourceSource: Grok Bot ChangelogGrok Voice Transcribe 2.0 released
Grok Voice Transcribe 2.0 is now available with higher accuracy on real-world audio and automatic language detection; xAI reports twice the accuracy of 1.0 at the same price. Batch and streaming, word timestamps, speaker separation, multichannel transcription for up to eight channels, and key term prompting are retained.
Usage details
Who can use it: Speech-to-Text API users; batch is $0.10/hour, streaming is $0.20/hour.
How to use it: Use `grok-voice-transcribe-2.0` in the Speech-to-Text API; to stay on the older version, explicitly pin the `grok-voice-transcribe-1.0` slug.
Tip: Enable key terms and diarization options for phone calls and short multilingual commands, then send the output directly into your workflow.
✓ Verified by official sourceModels: grok-voice-transcribe-2.0, grok-voice-transcribe-1.0Source: Introducing Grok Voice Transcribe 2.0Source: Release NotesQwen3.8-LiveTranslate separates speakers and translates in real time
Qwen3.8-LiveTranslate announced that its Interleave architecture reduced average latency from 2,8 seconds to 2,3 seconds. It adds real-time speaker separation, synchronized display of the source and translation, and long-context interpretation of proper nouns.
Usage details
Who can use it: Available through the API and demo; 60 languages are supported for input and 29 for audio output, with Turkish included in both lists.
How to use it: Use the qwen3.8-livetranslate-flash-realtime model in a DashScope real-time session; set the target language and optional audio output in session.update.
Tip: For meeting translations, retain the source text as well so you can verify names and technical terms later.
✓ Verified by official sourceModels: qwen3.8-livetranslate-flash-realtimeSource: Qwen3.8-LiveTranslate: Names the speaker. Carries the meaning.From audio-video understanding to agent actions with Qwen3.8-Omni-Flash
Qwen3.8-Omni-Flash processes text, images, audio, and video with a 1M-token context, focusing on task completion through reasoning, tool calling, and web search. Qwen also opened Qwen-MM-Plugins and Qwen-Live Harness, adding workflows such as long-form video/audio analysis, Video2Note, and Omni Skill Creator around the model.
Usage details
Who can use it: Qwen AI Platform and Model Studio API; Model Studio regions and plans are listed in separate documentation. The announcement reported that audio-visual input prices had dropped significantly compared with the previous version.
How to use it: Try the model in Qwen Studio or make a Chat Completions/Responses call to the Qwen3.8-Omni-Flash API; for long videos, use the agent’s evidence-gathering workflow.
Tip: For videos lasting hours, reduce unnecessary frame-processing costs by clearly stating the question and the evidence to look for first.
✓ Verified by official sourceModels: Qwen3.8-Omni-FlashSource: Qwen3.8-Omni-Flash: Omni Senses. Agentic Delivery.Source: Alibaba Cloud Model Studio: Qwen-OmniCompliance API adds support for Chrome sessions and the cache diagnostics field
The Compliance API’s native session endpoints now also return Claude in Chrome session transcripts. As of the same date, responses using the cache diagnostics beta header always include the diagnostics field; it returns null when there is no data.
Usage details
Who can use it: Claude Enterprise; a Compliance Access Key and the read:compliance_user_data scope are required. The cache diagnostics beta header is available to API users.
How to use it: Filter Compliance sessions with product_surface=claude_in_chrome; read the diagnostics field using the cache-diagnosis-2026-04-07 header.
Tip: When storing Chrome transcripts, separately review user privacy, access scope, and retention period.
✓ Verified by official sourceSource: Claude Platform API sürüm notlarıAPI access to Gemini 2.5 models limited to users with prior usage
To preserve capacity, Google has begun limiting access to Gemini 2.5 models to developers who have actively used them before. The models have not been deprecated and remain available in the API; Google recommends 3.5 Flash-Lite or 3.8 Flash for new projects.
Usage details
Who can use it: API users who have actively used Gemini 2.5 models before; access may be limited for new projects.
How to use it: For new projects, switch model selection to Gemini 3.5 Flash-Lite or Gemini 3.8 Flash; check access for existing 2.5 workflows.
Tip: Inventory dependencies on older models and run backward-compatibility, quality and cost tests with the new model.
✓ Verified by official sourceModels: Gemini 2.5 Pro, Gemini 2.5 Flash, Gemini 2.5 Flash-LiteSource: Release notes | Gemini API - Google AI for DevelopersSource: Gemini deprecationsAntigravity Agent 09-2026 release published
antigravity-preview-09-2026 replaces the previous 05-2026 release. For integrations that read only text output in a remote environment, changing the model string is sufficient; workflows using local tools require tool changes such as PascalCase parameters and line-range-based file editing.
Usage details
Who can use it: Developers using the Antigravity managed agent in the Gemini API Interactions API.
How to use it: Use antigravity-preview-09-2026 in the agent field; implement the new tool schema in clients that handle local tools and function_call.
Tip: Test remote and local operating modes separately before October 5, the sunset date for the 5-2026 release.
✓ Verified by official sourceModels: antigravity-preview-09-2026, antigravity-preview-05-2026Source: Release notes | Gemini API - Google AI for DevelopersSource: Gemini deprecationsKimi Code Desktop officially launches
The Desktop client, which brings Kimi Code’s Agent core to a graphical interface, has officially launched. It offers a workflow for opening projects, chatting with Agent, and monitoring tool calls and file changes in a single window.
Usage details
Who can use it: Initially available for macOS Apple Silicon/Intel and Windows; the Linux version followed on October 14.
How to use it: Download Kimi Code Desktop, open a project folder, and start an Agent session.
Tip: On first use, select the Always Ask or Ask When Needed permission mode to review file and command changes.
Notebooks in Gemini expands to schools and organizations
Gemini Notebook has expanded to Workspace accounts as a workspace that brings together topic-specific sources and conversations for students, educators and professionals. Users can add up to 10 sources to a notebook and generate a study guide, quiz or assignment submission from the same materials.
Usage details
Who can use it: All Google Workspace customers, Workspace Individual and personal Google accounts; an administrator must enable Gemini and Gemini Notebook. EEA access was added in the October 19 update.
How to use it: In the Gemini app, select New notebook from the left panel, then add a focus area and sources.
Tip: Create a separate notebook for each project or course to keep source context from getting mixed up.
✓ Verified by official sourceSource: Notebooks in Gemini: a dedicated workspace for focused, organized work, now for schools and organizationsTurn and active-time limits for Qwen Code Goal
Goal automation can now be limited by turn count and active time using model.goalMaxTurns and model.goalMaxActiveMinutes. When a limit is reached, a recovery turn is provided, usage is shown on the status card, and /goal resume opens a new window.
Usage details
Who can use it: Sessions using Qwen Code CLI, Web Shell, and Goal; the settings are disabled by default.
How to use it: Set model.goalMaxTurns and model.goalMaxActiveMinutes in settings; in Web Shell, approve the Goal suggestion.
Tip: For autonomous coding tasks, set token, turn, and time limits to reduce unexpected costs.
Qwen Code can delegate tasks to Claude Code and Codex
Qwen Code can assign subtasks to installed Claude Code or Codex and show permissions and results in the same conversation. Codex runs read-only by default; other ACP-compatible agents can also be added with an executor definition under .qwen/agents.
Usage details
Who can use it: macOS/Linux; Windows requires WSL. A Claude Code or Codex account and installation are required.
How to use it: Request a claude-code or codex subagent in the conversation; for a custom agent, define an executor under .qwen/agents.
Tip: Use Codex as a read-only review agent before asking it to make code changes.
Run Qwen Code workflows by name and set a per-turn token limit
Saved workflows can now be started directly in a conversation by name or with a slash command, and a per-message token cap can be set with an additive value such as +500k. Workflows can be distributed in extension packages; a preview of the fan-out and subagent structure is shown before execution.
Usage details
Who can use it: Developers using Qwen Code workflows and extensions.
How to use it: Use a slash command such as /audit for a saved workflow; add a limit like +500k to the end of the message.
Tip: For costly workflows, run a test with a low token cap first.
Qwen Code Web Shell adds live web preview and versioned HTML delivery
Web Shell lets users view a local development server in the right panel at desktop and mobile widths. Every HTML output delivered in a conversation is archived as a separate version and can be reopened later.
Usage details
Who can use it: Qwen Code Web Shell users.
How to use it: Open web preview in the right panel and enter the local development URL; select a version from the HTML delivery cards.
Tip: Check desktop and mobile views side by side before submitting a set of changes.
Qwen Code web_search session quota is now configurable
Web search is now limited to a default of 200 calls per session. This limit can be changed with the tools.webSearch.maxPerSession setting; when the quota is reached, the agent continues with the information it has gathered and reports that additional budget is needed.
Usage details
Who can use it: Sessions using Qwen Code’s web_search tool.
How to use it: Set the tools.webSearch.maxPerSession value in settings.
Tip: For long research tasks, combine queries and reduce duplicates before raising the search limit.
Browser calls and image previews are now shown in the transcript
Desktop 1.0.1 made browser tool calls visible in the transcript, shows links’ full URLs on hover, and added gallery navigation and a right-click menu to image previews.
Usage details
Who can use it: Kimi Code Desktop 1.0.1 users.
How to use it: Open a browser call or image card in the transcript; hover over a link to view its URL and actions.
Tip: For image-based code reviews, use gallery view to compare with previous screenshots.
✓ Verified by official sourceSource: Kimi Code Desktop ChangelogClaude Cowork and chat merge into one experience
Claude is beginning to combine simple questions and longer tasks such as reports, spreadsheets, and presentations in the same conversation, without asking users to choose Chat or Cowork mode. Chats, tasks, Projects, connectors, Skills, and settings are moving to the new experience.
Usage details
Who can use it: Gradual rollout; initially available on web, desktop, and mobile for Pro and Max. In Team beta and Enterprise, the Chat and Cowork unified setting must be enabled.
How to use it: In Claude, write the desired output directly; manage connected apps under Customize > Connectors and global instructions under Settings > Your Claude.
Tip: For tasks with real-world impact, check the Auto/Manual permission setting in the message box and approve critical steps.
✓ Verified by official sourceSource: Claude Cowork ve sohbet tek Claude olduSource: Claude uygulama sürüm notlarıDesign, Slides, and Docs can now be created in any conversation
Claude can now create a design, presentation, or living document in any conversation; the feature has also come to Claude Code and the Artifacts tab. Claude Design supports canvas editing and design system imports, Claude Slides supports presentation generation, and Claude Docs supports team-edited documents.
Usage details
Who can use it: Artifacts is available on all plans, including Free; Docs, Slides, and Design are in beta on Enterprise and remain off until an administrator enables them.
How to use it: Ask for a design, deck, or doc in a conversation, or select Output > the relevant template from the message box.
Tip: For enterprise visual consistency, first import your design system on the Artifacts page.
Grok Build gains project memory
Grok Build has added a memory system that remembers project rules, decisions, and persistent facts across sessions. After work is completed, it turns important information into topic notes in the background and reads them in later sessions before changing relevant code.
Usage details
Who can use it: Grok Build users; activates in new sessions.
How to use it: Start a new session to continue working with Grok Build; view notes with `/memory` and edit them with `/dream`.
Tip: Explicitly ask it to save project conventions and test commands to memory; don’t turn temporary task statuses into persistent notes.
✓ Verified by official sourceSource: Memory in Grok BuildWhatsApp Business setup with an agent is now possible
WhatsApp Business Tools MCP connects Service Provider AI coding agents to the WhatsApp Business Platform. Developers can use natural language to create accounts, phone numbers, and message templates; register for the Cloud API and complete OTP verification; and start configuring test messages and webhooks.
Usage details
Who can use it: Discoverable on Claude, Codex, and ChatGPT; it may not be available to everyone due to a phased rollout. It was released for development and testing workflows, not for production-scale messaging.
How to use it: Add WhatsApp Business Tools MCP to your agent’s MCP configuration and sign in with Facebook Login for Business.
Tip: During initial setup, have the agent check the Terms of Service, payment method, and Business Verification status in advance.
✓ Verified by official sourceSource: Announcing WhatsApp Business Tools MCP: Set up and manage WhatsApp business from your AI agentAPI key creation and expiration governance
Organization and project administrators can now allow only service account keys, only user-owned project keys, or disable new key creation entirely. Expiration dates can be added to project keys at creation, and a maximum key lifetime can be enforced at the organization or project level.
Usage details
Who can use it: OpenAI API organization and project administrators.
How to use it: Configure the API key creation governance and maximum key lifetime policies in Platform settings.
Tip: Apply short lifetimes and regular rotation to production keys; check that existing keys are not affected by these changes.
✓ Verified by official sourceSource: OpenAI API changelogGemini 3.8 Live models reach GA
Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking have been released as GA for low-latency voice applications. The first model is designed for fast dialogue and asynchronous background tool calls, while Extended Thinking is designed for deeper, multi-step reasoning while continuing the conversation.
Usage details
Who can use it: Gemini API and Google AI Studio; on the product side, Gemini Live, Search Live and select Workspace experiences. Plan, language, age and regional restrictions vary by product.
How to use it: Select the gemini-3.8-live or gemini-3.8-live-extended-thinking model in the Gemini Live API guide.
Tip: Use Live for low-latency customer conversations and Extended Thinking for complex background tasks.
✓ Verified by official sourceModels: Gemini 3.8 Live, Gemini 3.8 Live Extended ThinkingSource: Release notes | Gemini API - Google AI for DevelopersSource: Introducing Gemini 3.8 Live and 3.8 Live Extended ThinkingMeta One subscriptions bring higher AI usage limits
Meta announced Meta One plans comprising subscriptions to individual products and bundles for Instagram, Facebook, WhatsApp, and Meta AI. Core and Premium offer more image and video generation with Meta AI and higher usage limits for in-app AI tools, while business and creator plans expand access to Meta Business Agent.
Usage details
Who can use it: Plans are available globally; individual product plans start at 2,99 US dollars per month, individual bundles at 7,99 US dollars, and creator/business bundles at 14,99 US dollars. Pricing, benefits, and availability may vary by country, app, and account.
How to use it: Open Meta One or the subscription flow in the relevant app and choose one of the Core, Premium, Essential, Advanced, Expert, or Max options listed on your account.
Tip: If you only need a higher Meta AI generation limit, compare the Core or Premium plan instead of business tools.
✓ Verified by official sourceModels: MuseSource: Introducing Meta One: A Subscription Service With More Features and AI to Create, Connect, and Stand OutSalesforce in Claude beta released
Developed with Salesforce, the plugin brings sellers’ accounts, opportunities, and sales pipeline into Claude. Its 37 ready-made sales Skills provide call preparation, opportunity reviews, a pipeline dashboard, and CRM updates after approval; the Slack connector also covers deal channels.
Usage details
Who can use it: Beta on all paid Claude plans; requires approval through Salesforce’s beta program and setup by an organization administrator.
How to use it: Have an administrator request access through AgentExchange, connect Salesforce for the organization, and install the plugin from Claude Marketplace.
Tip: Keep the default approval step for CRM writes and limit automatic updates to low-risk fields.
✓ Verified by official sourceSource: Salesforce’u Claude’a getirmeSource: Claude uygulama sürüm notlarıMCP-based third-party connectors launch in Workspace
Gemini in Workspace now works with Asana, Atlassian Rovo, HubSpot, Intuit Mailchimp, Intuit QuickBooks, Monday and Salesforce through Model Context Protocol integrations. Users can access information in these services from Docs, Sheets, Slides and Chat without switching tabs.
Usage details
Who can use it: Eligible Business and Enterprise plans, Google AI, Pro and Ultra, Enterprise Essentials Plus and certain education add-ons; administrator activation is required.
How to use it: Manage third-party connectors in the admin console at Apps > Google Workspace Marketplace apps > Apps list; users can access them from the Gemini side panel in Workspace.
Tip: Enable MCP connections by OU or group using a least-privilege policy.
✓ Verified by official sourceSource: Connect to more tools with Gemini in Google WorkspaceBots can upload files to Drive and OneDrive and attach them to Gmail
Grok Bot can now upload files to Google Drive or OneDrive and attach them to Gmail drafts. This makes it easier to move reports and files generated by the Bot into connected workspaces without leaving the chat.
Usage details
Who can use it: Grok Bot users with a connected Google Drive, OneDrive, or Gmail account.
How to use it: Connect the relevant app through Connect Apps, then ask the Bot to upload the file or attach it to a Gmail draft.
Tip: Test corporate folder permissions and sharing scope with a low-sensitivity file first.
✓ Verified by official sourceSource: Grok Bot ChangelogRoutines can be paused and resumed
Grok Bot’s scheduled routines can now be paused without deleting them and resumed later from where they left off. This makes it possible to suspend seasonal or temporary workflows without setting them up again.
Usage details
Who can use it: Grok Bot accounts that can use Routines.
How to use it: In the Routines section of Bot details, right-click the relevant routine and select Pause or Resume.
Tip: Pause a routine instead of deleting it during temporary maintenance periods; check connected connector sessions before enabling it again.
✓ Verified by official sourceSource: Grok Bot ChangelogOn-demand conversation compaction enters beta in the Messages API
The Messages API can now summarize a conversation whenever the developer chooses, using the compaction parameter. The API returns a signed compaction block; this block can replace older messages in subsequent requests while preserving the latest turns verbatim.
Usage details
Who can use it: Claude API beta; requires the compact-2026-09-04 beta header.
How to use it: Send the compaction parameter at the top level of the request, then use the returned signed compaction block as the first block in the next request.
Tip: To preserve short-term context, leave the latest user and tool turns intact and compact only the older section.
✓ Verified by official sourceSource: Claude Platform API sürüm notlarıSend files to Kimi from the operating system menu
Work 3.2.9 added Share to Kimi and Open with Kimi workflows for supported file types, allowing files to be opened from other apps’ share menus or right-click menus.
Usage details
Who can use it: Kimi Work 3.2.9 users; supported file types and operating system menus.
How to use it: For a file, select Share to Kimi from the share menu or Open with Kimi from the right-click menu.
Tip: Use the operating system’s share workflow directly instead of copying the file to the Work folder first.
✓ Verified by official sourceSource: Kimi Work Release NotesPower mode keeps tasks running when the screen is off
Work 3.2.9 added the Keep Awake After Screen Off option, which allows the screen to turn off while keeping the system and background tasks running with Remote Control.
Usage details
Who can use it: Kimi Work 3.2.9 users.
How to use it: Turn on Keep Awake After Screen Off in power settings.
Tip: Use it to turn off the screen while long tasks continue; monitor power consumption on portable devices.
✓ Verified by official sourceSource: Kimi Work Release NotesPDF viewer gets a contents and page-jump panel
Work 3.2.9 updated the PDF viewer with a new index panel, enabling quick navigation from the table of contents and by page number.
Usage details
Who can use it: Kimi Work 3.2.9 users.
How to use it: Open a PDF, then select a section or page number from the viewer’s index panel.
Tip: In long reports, use the page field to jump directly to a page number provided by Agent.
✓ Verified by official sourceSource: Kimi Work Release NotesMismatch diagnostics added for Preserved thinking
Under the thinking-binding-controls beta header, the input_transformations response now includes the thinking_mismatch_allowed type. This record makes visible cases where a request was not rejected even though the thinking block failed the prefix check, and helps you monitor history edits in production.
Usage details
Who can use it: Claude API beta; especially for Claude Fable 5.1 and new accounts using preserved thinking.
How to use it: Send the thinking-binding-controls-2026-08-01 header and log thinking_mismatch_allowed records in input_transformations.
Tip: Before moving to enforcement, measure which client or history edit each mismatch record came from.
Desktop pet updated with a quick-reply bubble
Work 3.2.9 updated the desktop pet interface and made it possible to send quick replies from the speech bubble without opening the main window.
Usage details
Who can use it: Kimi Work users; the desktop pet feature must be enabled.
How to use it: Enable the desktop pet and open its speech bubble; send a quick reply directly from the bubble.
Tip: Use the pet for brief reminders and commands, and the main Work window for longer file tasks.
✓ Verified by official sourceSource: Kimi Work Release NotesChoose the role of an image in a video
Imagine now lets you choose the role an added image plays in video generation: First frame, Last frame, Loop, or Reference. In the same release, connection cards began appearing directly in chat for apps that require a connector.
Usage details
Who can use it: Grok Imagine users; video generation limits and plan entitlements apply.
How to use it: After adding an image in the Imagine creator, choose its role: First frame, Last frame, Loop, or Reference.
Tip: Try the starting/ending frame to control camera movement, or Reference mode for character consistency.
Bot can route web traffic through the user’s computer
The Route egress through this desktop option added to Settings lets the Bot’s computer send web traffic through the user’s own desktop connection instead of the cloud. This provides a new control for corporate network access and IP-based permissions.
Usage details
Who can use it: Accounts using the Grok Bot desktop app; team admins can restrict this with network controls.
How to use it: Turn on Route egress through this desktop in Settings → Computer.
Tip: Before enabling this setting, test local network permissions and verify that the Bot can’t run when the computer is off.
✓ Verified by official sourceSource: Grok Bot ChangelogGitHub, GitLab, Bitbucket, and Azure DevOps access
When the bot can’t access a repository, Grok Bot now displays a card suggesting a connection to GitHub, GitLab, Bitbucket, or Azure DevOps, then continues the task once access is granted. File paths on the bot’s computer can also be opened from chat.
Usage details
Who can use it: Grok Bot users; subject to the relevant repository and provider permissions.
How to use it: When the bot requests access, select a provider from the repository connection card, sign in, and allow the task to continue.
Tip: Grant the bot access only to the required repository and read/write scope rather than opening access to your entire organization.
✓ Verified by official sourceSource: Grok Bot ChangelogK2.8 Preview fully rolled out in Kimi Code
K2.8 Preview has been fully rolled out in Kimi Code, offering more efficient thinking than K2.7 Code and coding/Agent performance approaching K3. The model ID is unchanged and remains kimi-for-coding.
Usage details
Who can use it: 1 million token context across all Kimi Code membership tiers; thinking can be set to low, high, or max.
How to use it: Use the kimi-for-coding model ID in Kimi Code; set thinking effort to low, high, or max.
Tip: Start a new session and verify K2.8 Preview behavior before changing existing integrations.
Plugin added to deploy local web projects in one step
Work 3.2.7 introduced the website deployment plugin, which lets users deploy local website projects to the cloud in one step after installing the plugin.
Usage details
Who can use it: Kimi Work users; the website deployment plugin must be installed.
How to use it: Install the website deployment plugin from the plugin marketplace, then select the deployment action in the project task.
Tip: Before going live, check the local preview and the files that will be shared in the production environment.
✓ Verified by official sourceSource: Kimi Work Release NotesDeepSeek-V4.1-Flash available through GreenPT’s EU-hosted route
GreenPT announced that it was offering DeepSeek-V4.1-Flash on EU infrastructure shortly after its official launch on September 10. This is a third-party OpenAI-compatible access route hosted in Europe, separate from DeepSeek’s own API.
Usage details
Who can use it: European users with a GreenPT account; the provider’s EU hosting and pricing terms apply.
How to use it: In the GreenPT API, set the base URL to https://api.greenpt.ai/v1 and send the model name as deepseek-v4.1-flash.
Tip: Before enterprise use, confirm the data processing location, retention policy, and image/long-context behavior with the provider.
✓ Verified by official sourceModels: DeepSeek-V4.1-FlashSource: DeepSeek V4.1 Flash: GreenPT’de erken AB erişimiSource: DeepSeek-V4.1-Flash resmî duyurusuOCR for screenshots and desktop pinning added
Work 3.2.7 improved long screenshots in Launcher, and added one-click OCR in the screenshot editor and the option to pin a screenshot to the desktop as an always-on-top window.
Usage details
Who can use it: Kimi Work 3.2.7 users.
How to use it: Take a screenshot from Launcher, then click the OCR button in the editor or pin the image to the desktop.
Tip: Send the OCR result to Agent as context instead of typing out screenshot text by hand.
✓ Verified by official sourceSource: Kimi Work Release NotesDrag files and folders anywhere in the window
Work 3.2.7 added support for adding files and folders by dragging them anywhere in the app window.
Usage details
Who can use it: Kimi Work users.
How to use it: Drag a file or folder from Finder/Explorer onto the Work window.
Tip: Add a folder directly instead of selecting multiple files in it one by one.
✓ Verified by official sourceSource: Kimi Work Release NotesNew Work conversations now inherit settings automatically
Work 3.2.7 enables new conversations to automatically carry over conversation settings.
Usage details
Who can use it: Kimi Work users.
How to use it: Configure conversation settings once, then check that they’re applied automatically in a new Work conversation.
Tip: Use this inheritance behavior to standardize model, permission, and display settings for recurring tasks.
✓ Verified by official sourceSource: Kimi Work Release NotesAgents API released as a public beta
The Agents API makes Codex’s agent runtime available to developers, offering persistent sessions, streaming progress, context compaction, tool search, MCP, and sandbox options hosted by OpenAI or on your own infrastructure. Multi-agent support enables independent subtasks to run in parallel.
Usage details
Who can use it: Public beta for all developers; there is no additional Agents API fee, and tokens and tools used are billed separately.
How to use it: Follow the Agents API quickstart and define the model, tools, environment, and task in a single agent session request.
Tip: For long sessions, write task status and persistent files to the sandbox to make it easier to continue work after context compaction.
DeepSeek-V4.1-Flash open-weight release launched
DeepSeek launched the open-weight release of V4.1-Flash by publishing the Hugging Face model and technical report. The model card documents the MIT license, multimodal input support, and Transformers, vLLM, and SGLang workflows for local inference.
Usage details
Who can use it: Developers with Hugging Face access; local or compatible hosted use under the MIT license.
How to use it: Pull the deepseek-ai/DeepSeek-V4.1-Flash model from Hugging Face and follow the model card’s Transformers, vLLM, or SGLang instructions.
Tip: Given the model’s substantial hardware requirements, prototype with the provider API first, then calculate the cost of local deployment.
✓ Verified by official sourceModels: DeepSeek-V4.1-FlashSource: DeepSeek-V4.1-Flash resmî duyurusuSource: DeepSeek-V4.1-Flash model kartıDeepSeek-V4.1-Flash released
DeepSeek has released DeepSeek-V4.1-Flash, the smallest model in its new architecture family. The model was introduced with a 552B-parameter MoE structure, a new Causal Encoder–Decoder architecture, and a focus on higher speed and throughput.
Usage details
Who can use it: Available through the DeepSeek API, official web/app channels, and open-weight distribution.
How to use it: Use deepseek-flash as the model name in the API; for local deployment, follow the inference instructions on the Hugging Face model card.
Tip: For long agent tasks, measure lower active parameter usage and total task duration against previous V4 models.
✓ Verified by official sourceModels: DeepSeek-V4.1-FlashSource: DeepSeek-V4.1-Flash resmî duyurusuSource: DeepSeek-V4.1-Flash API duyurusuSource: DeepSeek-V4.1-Flash model kartıGemini Windows desktop app released
The Gemini app is now globally available on Windows 10 and 11. The Alt + Space shortcut, a separate workspace, Gmail and Drive context, Gemini Spark tasks, Nano Banana images and Gemini Omni videos are now available on desktop.
Usage details
Who can use it: Users of Windows 10 and 11; a Google AI subscription, age and regional requirements apply to Spark, image and video features.
How to use it: Download the Windows app from gemini.google/desktop; launch it with Alt + Space.
Tip: Test the shortcut to ensure it does not conflict with a key combination you use frequently, and review file permissions.
✓ Verified by official sourceModels: Gemini Spark, Nano Banana, Gemini OmniSource: The Gemini app is now available for WindowsSource: Gemini Apps' release updates & improvementsGPT-Live 1 API generally available
GPT-Live 1 is generally available in the API for full-duplex voice conversations that can continue while a backend model or agent uses tools. Voice sessions are billed by the second, and backend model and tool usage are billed separately.
Usage details
Who can use it: OpenAI API customers; voice sessions cost $0.05/minute, with backend usage billed separately.
How to use it: Start a full-duplex voice session using the Live sessions API or GPT-Live guide.
Tip: Track voice session and backend agent costs with separate metrics; end the session when the user is inactive.
Auto decision added to Managed Agents permission policies
Managed Agents permission policies can now evaluate each agent or MCP tool call on the server and execute it, deny it, or require user approval. The agent.tool_use and agent.mcp_tool_use events report the evaluation result in the evaluation field.
Usage details
Who can use it: Claude Managed Agents; server-side permission policies.
How to use it: Add an auto rule to the permission policy and process the evaluation and evaluated_permission fields in the agent event stream.
Tip: For tools that can write data or incur costs, start with rules that require approval rather than auto.
✓ Verified by official sourceSource: Claude Platform API sürüm notlarıQwen Code Channel bots add multiple named tasks
Multiple named tasks can now run concurrently in the same group across DingTalk, Feishu, and other Channel bots. Each task has a separate history and optional worktree, allowing PR reviews and feature development to happen in the same group chat without file conflicts.
Usage details
Who can use it: Qwen Code teams using Channel integration; requires sessionScope=user and multiSession=true.
How to use it: Create a task with /session new <name>, switch to it with /session use <name>; add --worktree for file isolation.
Tip: Name one task for code review and another for implementation, and run each in a separate worktree.
✓ Verified by official sourceSource: Qwen Code Weekly: Look Back on Workflow Runs, Pair Plan With YOLO, Multitask in Group Chatweb_search is enabled automatically in Qwen Code with a Bailian key
The built-in web_search becomes available automatically when using a Bailian Standard API Key or Token Plan. With Coding Plan, provider configuration must be specified manually, and search costs are charged to the account used.
Usage details
Who can use it: Bailian Standard API Key or Token Plan users; Coding Plan requires additional configuration.
How to use it: Start a session with Standard API Key or Token Plan authentication; for Coding Plan, configure the Bailian provider.
Tip: Track web search charges to the key you use separately in your team budget.
✓ Verified by official sourceSource: Qwen Code Weekly: Look Back on Workflow Runs, Pair Plan With YOLO, Multitask in Group ChatV4 Pro API service continued after September 14
DeepSeek’s changelog announced that, in response to user demand, the DeepSeek V4 Pro API service would continue after September 14, 2026. It stated that the billing method would not change as a result of this decision.
Usage details
Who can use it: DeepSeek API users; the continuation of V4 Pro service was announced in the official changelog.
How to use it: Continue using model=deepseek-v4-pro in the API; check the Models & Pricing page for current pricing and version information.
Tip: Compare V4 Pro and V4.1-Flash outputs on the same task set to reassess the quality-cost trade-off.
V4-Flash and V4-Flash-Vision-Exp retired
DeepSeek retired the previous-generation V4-Flash and V4-Flash-Vision-Exp API models. For backward compatibility, the deepseek-v4-flash and deepseek-v4-flash-vision-exp names were temporarily redirected to V4.1-Flash.
Usage details
Who can use it: Old model names work through a temporary compatibility route; use deepseek-flash for new integrations.
How to use it: Replace old model names in your code with deepseek-flash; during the transition, log which version the old names route to.
Tip: Don’t rely on old names permanently; update the model name explicitly to reduce behavioral differences after retirement.
✓ Verified by official sourceModels: DeepSeek-V4-Flash, DeepSeek-V4-Flash-Vision-Exp, DeepSeek-V4.1-FlashSource: DeepSeek-V4.1-Flash API duyurusuSource: Models & PricingV4.1-Flash supports 1M-token context
The V4.1-Flash model card specifies context of up to 1 million tokens with image and text inputs, and a recommended output limit of at least 256K. This makes it easier to keep long code repositories, documents, and multi-step agent sessions in a single context.
Usage details
Who can use it: Developers using the open-weight model or the V4.1-Flash route in the DeepSeek API; actual limits may vary by provider and client.
How to use it: Use the model card’s 1M context_window value as a guide to send long documents in one request or in controlled sections.
Tip: Control input costs with relevance ranking and caching instead of filling the 1M context completely.
✓ Verified by official sourceModels: DeepSeek-V4.1-FlashSource: DeepSeek-V4.1-Flash model kartıSource: DeepSeek-V4.1-Flash resmî duyurusuNative image understanding comes to the V4.1-Flash API
DeepSeek-V4.1-Flash went live in the API with native multimodal support that processes text and images in the same model. This enables visual question answering, document tasks, and image-based agent workflows without requiring a separate experimental vision model.
Usage details
Who can use it: Accounts using deepseek-flash in the DeepSeek API; V4.1-Flash image input support applies.
How to use it: Use model=deepseek-flash in a Chat Completions request and add text and image_url blocks to the user content.
Tip: When testing image and text prompts together, separately record image size, OCR quality, and token budget.
✓ Verified by official sourceModels: DeepSeek-V4.1-FlashSource: DeepSeek-V4.1-Flash API duyurusuSource: DeepSeek-V4.1-Flash model kartıAPI prices cut with V4.1-Flash
DeepSeek cut API prices with V4.1-Flash’s more efficient architecture and retained the peak/off-peak model. The new prices took effect on September 10, 2026, at 04:00 UTC; off-peak prices remained 50 percent of peak prices.
Usage details
Who can use it: DeepSeek API users; peak/off-peak hours and current rates are listed on the Models & Pricing page.
How to use it: Schedule flexible and batch jobs for weekends or off-peak hours; use the current API pricing table to check prices.
Tip: Build a real-cost dashboard based on cache hits, input/output tokens, and the UTC time zone.
✓ Verified by official sourceModels: DeepSeek-V4.1-FlashSource: DeepSeek-V4.1-Flash resmî duyurusuSource: DeepSeek-V4.1-Flash API duyurusuSource: Models & PricingConnect to a Managed Agents session from the ant CLI
ant CLI 1.32.0 adds the ant beta:sessions connect command. The terminal can now monitor a Managed Agents session live, send messages, and approve or deny tool calls awaiting approval.
Usage details
Who can use it: ant CLI 1.32.0 and access to Claude Managed Agents.
How to use it: Run the ant beta:sessions connect command; use the --web option to open the Claude Console session viewer locally.
Tip: For long-running agent tasks, use the terminal connection for monitoring and approvals; don’t print secrets to terminal history.
✓ Verified by official sourceSource: Claude Platform API sürüm notlarıMem0 external memory integration prepared for Qwen Code
An official extension package provides an installation and configuration flow for Mem0. The integration brings the ability to read, write, delete, and manage project or personal memories to Qwen Code, along with automatic recall when enabled by an administrator.
Usage details
Who can use it: Qwen Code users with a Mem0 service and API access.
How to use it: Install the official Mem0 extension and configure the connection; enable the administrator option for automatic recall.
Tip: Keep personal and project memories separate to prevent sensitive notes from being shared with the wrong project.
✓ Verified by official sourceSource: Qwen Code Weekly: Look Back on Workflow Runs, Pair Plan With YOLO, Multitask in Group ChatQwen Code runs Plan and YOLO modes together
Plan now works as a separate toggle from permission mode. This lets you enable planning together with Default, Auto Edit, Auto, or YOLO; Plan turns off automatically after approval.
Usage details
Who can use it: Qwen Code Web Shell and relevant CLI users.
How to use it: In Web Shell, enable the Plan toggle and your desired execution mode separately.
Tip: For risky projects, review the plan with Plan + Auto Edit first, then switch to full YOLO.
✓ Verified by official sourceSource: Qwen Code Weekly: Look Back on Workflow Runs, Pair Plan With YOLO, Multitask in Group ChatQwen Code gets a workflow management page
A dedicated Workflows page with Saved, Running, and History tabs has been added to Web Shell. Users can view progress, created agents, pending permissions, and token usage, and stop or rerun workflows.
Usage details
Who can use it: Qwen Code v0.23.1–v0.23.2 Web Shell users.
How to use it: Open Workflows from the left menu or run /workflows in a conversation.
Tip: Keep frequently used control workflows in the Saved tab and compare past runs by cost.
✓ Verified by official sourceSource: Qwen Code Weekly: Look Back on Workflow Runs, Pair Plan With YOLO, Multitask in Group ChatV4.1-Flash reduces KV cache consumption
DeepSeek said V4.1-Flash reduces the HBM required for KV cache to one-quarter and SSD storage to one-eighth of the previous generation. The architecture aims to reduce cache costs and infrastructure pressure, particularly in input-heavy agent workloads.
Usage details
Who can use it: Part of the model architecture in DeepSeek API and V4.1-Flash open-weight/inference deployments.
How to use it: Use deepseek-flash in the API; in your own deployment, enable KV cache support in a V4.1-Flash-compatible inference engine.
Tip: Keep repeated system prompts and tool definitions fixed, and separately monitor the cache hit rate.
✓ Verified by official sourceModels: DeepSeek-V4.1-FlashSource: DeepSeek-V4.1-Flash resmî duyurusuSource: DeepSeek-V4.1-Flash API duyurusuV4.1-Flash adds a continuous reasoning effort setting
The V4.1-Flash model card describes support for continuously adjustable reasoning effort from 1–100, letting users choose between accuracy and inference cost. Higher effort aims to improve quality on difficult tasks at the cost of longer thinking and higher costs.
Usage details
Who can use it: Especially in open-weight and compatible inference/agent environments; the API interface may also offer named low/high/max options.
How to use it: In a local or compatible inference layer, set reasoning_effort to a value from 1–100 based on task difficulty.
Tip: Use A/B testing to select low effort for simple classification and high effort for coding and tool-using tasks.
✓ Verified by official sourceModels: DeepSeek-V4.1-FlashSource: DeepSeek-V4.1-Flash model kartıSource: DeepSeek Thinking Mode rehberiSupport for WorkBuddy, CodeBuddy, and OpenCode added
DeepSeek announced that V4.1-Flash is fully supported by WorkBuddy, CodeBuddy, and OpenCode. These integrations make it easier to add the model to existing coding and agent workflows without writing a custom connector.
Usage details
Who can use it: DeepSeek API customers using the relevant third-party coding/agent tools.
How to use it: Select the DeepSeek provider in your tool, enter your API key, and use deepseek-flash as the model.
Tip: Before production, verify each tool’s support for the Responses API, image input, and tool calls separately.
✓ Verified by official sourceModels: DeepSeek-V4.1-FlashSource: DeepSeek-V4.1-Flash resmî duyurusuSource: DeepSeek-V4.1-Flash API duyurusuWeekly usage and monthly optional spending limit
The Usage & Billing screen now shows weekly usage and the plan, warns in chat when usage approaches the weekly limit, and lets users set a monthly on-demand spending limit. This makes Bot usage costs more visible to both users and administrators.
Usage details
Who can use it: Available on Grok Bot plans; optional usage and administrator limits are subject to plan policy.
How to use it: Go to Settings → Usage & Billing to monitor weekly usage and set a monthly on-demand limit.
Tip: Set a monthly limit before enabling automated routines, and monitor usage spikes from intensive workflows.
✓ Verified by official sourceSource: Grok Bot ChangelogRemote Control exits experimental status and becomes generally available
Kimi Code CLI 0.42.0 made Remote Control generally available without requiring an experimental flag. The feature lets you take over a local session from a phone or another computer.
Usage details
Who can use it: Kimi Code users; available through a connected device or phone.
How to use it: Start a Remote Control connection in Kimi Code, then open the connection or QR code on the other device.
Tip: Open remote access connections only on trusted devices, and close them when you’re done.
✓ Verified by official sourceSource: What's New | Kimi Code DocsMultiple accounts and Bot Marketplace
Multiple accounts can now be added and switched between from the menu in Grok Bot desktop; each account has its own Bots and data. Marketplace also now lets users discover plugins and Bots on the same home page, with categories and search.
Usage details
Who can use it: Grok Bot desktop users; Marketplace content may vary based on account and team access.
How to use it: Add an account from the Account menu, then select and add a plugin or Bot from Marketplace.
Tip: To avoid mixing personal and work accounts, check each account’s connector and Bot lists separately.
✓ Verified by official sourceSource: Grok Bot ChangelogSortable media previews and new image formats added to the web interface
Kimi Code 0.42.0 added a sortable preview bar to make it easier to reference attachments in text; it also added support for HEIC, HEIF, and BMP images.
Usage details
Who can use it: Users of the Kimi Code Web interface.
How to use it: Attach an image or video to the message box and change its order in the preview bar; attach a supported HEIC/HEIF/BMP file directly.
Tip: Put multiple screenshots in the correct order to prevent Agent from mixing up their visual context.
✓ Verified by official sourceSource: What's New | Kimi Code DocsCache diagnostics fingerprint storage behavior clarified
When the Cache diagnostics beta header is sent by itself, the API no longer stores the request fingerprint. To retain the fingerprint and the subsequent previous_message_id relationship, the diagnostics object must be sent on every turn, and previous_message_id must be null on the first turn.
Usage details
Who can use it: Claude API; integrations using the cache-diagnosis-2026-04-07 beta header.
How to use it: Send diagnostics:{previous_message_id:null} in the first request, then send a diagnostics object containing the previous message ID in subsequent requests.
Tip: Before enabling cache diagnostic logs in production, review the retention policy for sensitive request fingerprints.
✓ Verified by official sourceSource: Claude Platform API sürüm notlarıChatGPT Images 2.5 and GPT Image 2.5 API models
ChatGPT Images 2.5 brings sharper details, more precise editing, and faster generation, along with workflows for templates, mobile sketches, comments, and prompt sharing. On the same day, Sunburst, focused on precise editing, and Flare, focused on fast everyday generation, launched in the API; both models support the `xhigh` and `max` quality settings.
Usage details
Who can use it: ChatGPT Images 2.5 is subject to existing image generation limits in ChatGPT; GPT Image 2.5 Sunburst/Flare is available to developers using the Image API and the Responses API image generation tool.
How to use it: In ChatGPT, open Images > Templates or select Sketch with `@` on mobile; in the API, use the relevant image model ID and quality setting.
Tip: For a precise edit, start with Sunburst; for fast, high-quality everyday images, start with Flare and compare using the same prompt.
✓ Verified by official sourceModels: ChatGPT Images 2.5, GPT Image 2.5 Sunburst, GPT Image 2.5 Flare, gpt-image-2.5-sunburst, gpt-image-2.5-flareSource: ChatGPT yayın notlarıSource: OpenAI API changelogEmail and Slack drafts with administrator Auto-review
Bots can now prepare a Gmail email or Slack message as a draft card and wait until the user selects Send or Discard. Enterprise administrators can require Auto-review and add team-wide rules.
Usage details
Who can use it: The draft feature is available to Grok Bot users; mandatory Auto-review is available to Enterprise administrators.
How to use it: Ask the bot to prepare an email or Slack message; an Enterprise administrator can enable Auto-review and team rules in the dashboard.
Tip: Keep draft mode on for all messages sent to external customers, and automate only trusted internal notifications.
✓ Verified by official sourceSource: Grok Bot ChangelogGPT-Rosalind is generally available through the trusted-access program
GPT-Rosalind is generally available through the trusted-access program for approved internal life sciences research. Standard pricing was announced at $5 per 1 million input tokens, $0.50 per 1 million cached input tokens, and $25 per 1 million output tokens; billing began on October 5.
Usage details
Who can use it: Approved life sciences researchers and API organizations; subject to trusted-access approval.
How to use it: Apply to the trusted-access program and, once approved, use the `gpt-rosalind-research` model in the API.
Tip: Before submitting research data, check access approval, data retention terms, and the scope of sensitive data.
✓ Verified by official sourceModels: GPT-Rosalind, gpt-rosalind-researchSource: OpenAI API changelogMuse introduced as a personal AI agent
Meta introduced Muse, which takes on tasks such as sending emails, planning trips, filling out forms, and turning long-term goals into plans, rather than simply providing answers. Muse can keep working after the user closes it, moving between chats, the browser, and connected services while requesting approval for sensitive steps.
Usage details
Who can use it: Available in the US on iOS, Android, and muse.ai; support for AI glasses is coming later. Most use is free, with subscription plans for more.
How to use it: Write your goal in natural language in the Muse app or on WhatsApp; select the apps you want to connect and follow the approval flow before actions are taken.
Tip: For your first task, clearly specify the budget, deadline, and which steps require approval.
✓ Verified by official sourceModels: Muse SparkSource: Introducing Muse: The World’s First Personal AI Agent Built for EveryoneMuse Secure VM and user control settings arrive
Muse was introduced with the Muse Secure VM architecture, which runs on a separate cloud computer for each user. Users can choose which apps to connect and set permissions such as read and send; the Sentinel layer controls internet access, an activity log is shown, users can turn off training, and they can ask Muse to forget specific memories.
Usage details
Who can use it: Available on accounts where Muse is offered; initially in the US, on iOS, Android, and web. Meta said it will also offer Confidential VM later in the year, encrypted with a key held only by the user.
How to use it: Manage connected service permissions, data training preferences, and remembered information in Muse settings; use the confirmation screen for sensitive actions.
Tip: For email, grant read-only permission first; enable sending and purchasing permissions separately when needed.
✓ Verified by official sourceModels: Muse SparkSource: Introducing Muse: The World’s First Personal AI Agent Built for EveryonePrompt Cache Diagnostics is now generally available in the Responses API
Prompt Cache Diagnostics lets you compare cache reuse with the previous response and diagnose cache miss causes for GPT-5.6 and later supported models. It provides measurable feedback to improve cache hit rates in long-context agents.
Usage details
Who can use it: API customers using GPT-5.6 and later supported models in the Responses API.
How to use it: Enable the diagnostic fields and dashboard described in the Prompt Cache Diagnostics guide in the Responses API.
Tip: First check that system instructions and tool order remain unchanged; move variable parts to the end of the prompt.
✓ Verified by official sourceSource: OpenAI API changelogPlugin installation from GitHub and MCP status management added
Work 3.2.6 lets users install shared plugins through a GitHub connection and view and manage MCP connection status on the plugin details page.
Usage details
Who can use it: Kimi Work plugin marketplace users.
How to use it: Add a GitHub connection in the plugin marketplace; use the MCP connection status and connection management controls on the details page.
Tip: Review the source code and requested MCP permissions before installing external GitHub plugins.
✓ Verified by official sourceSource: Kimi Work Release NotesTask-based plugin recommendations and installation flow added
Work 3.2.6 now recommends plugins suited to a task in the workflow and installs them with user approval.
Usage details
Who can use it: Kimi Work users; installation of a recommended plugin requires approval.
How to use it: Start a task; when a suitable plugin is recommended, review its details and approve the installation.
Tip: Don’t approve automatic installation before reviewing plugin permissions and MCP connection status.
✓ Verified by official sourceSource: Kimi Work Release NotesIn-chat forms and AutoFill
When the bot needs information such as a login, it can open a form in chat and fill out the page instead of handing control over to the user’s computer screen. An AutoFill shortcut for text fields has also been added on Mac.
Usage details
Who can use it: Users with Grok Bot access; site and field support may vary by page.
How to use it: Fill out the chat form when the bot requests information; on Mac, right-click a text field and use AutoFill.
Tip: Use Secure Form or AutoFill fields instead of sharing passwords and payment codes in chat messages.
✓ Verified by official sourceSource: Grok Bot ChangelogImagine moves into Projects, with two new voices
Images can now be generated directly from the composer in Grok Projects; Imagine has been integrated into the project workspace. The same release adds two new voices, Aurora and Liora, to Voice Library.
Usage details
Who can use it: Grok web users; voice options and Imagine generation are subject to plan limits.
How to use it: Open a Project and select Imagine from the composer; for voice use, select Aurora or Liora from Voice Library.
Tip: When generating images from project context, keep files and instructions in the same Project to avoid repeatedly transferring context.
✓ Verified by official sourceSource: Grok Release NotesNew Library page brings together all files and creations
A new Library page is available for content uploaded to Grok or generated by Grok. It brings grid and table views, sorting, visibility filters, and previews for documents, PDFs, Markdown, SVG, and archives together in one place.
Usage details
Who can use it: Grok web users; content visibility and team-sharing settings apply.
How to use it: Open Library from the Grok sidebar, then choose a view, sorting, and filters.
Tip: Separate long-term project files using tags and visibility filters to make them quicker to reuse.
✓ Verified by official sourceSource: Grok Release NotesRemote control of the Kimi Work desktop from a phone is now available
Work 3.2.5 added the Remote Control feature, which lets you control Kimi Work on your desktop from your phone.
Usage details
Who can use it: Kimi Work 3.2.5 users; a desktop and phone connection is required.
How to use it: Start Remote Control in Kimi Work and connect your phone.
Tip: Use it to keep track of long-running Agent tasks without staying at your desk.
✓ Verified by official sourceSource: Kimi Work Release NotesApps feature makes Agent’s web operations visible
Work 3.2.5 added the Apps feature, which makes Agent’s web operations and results more visible.
Usage details
Who can use it: Kimi Work users.
How to use it: During an Agent task, open Apps to review the results of web operations.
Tip: For web-based tasks, verify which application steps Agent carried out in the results card.
✓ Verified by official sourceSource: Kimi Work Release NotesLyria 3.5 comes to the Gemini app
Lyria 3.5 is now available in the Gemini app, with more expressive vocals and richer arrangements. Users can choose a genre or vocal/instrumental format, start from templates and generate short or long tracks.
Usage details
Who can use it: Globally available to all users on Gemini web and mobile; also available in Google AI Studio, Flow Music and Google Vids.
How to use it: Open the music area in Gemini and specify the genre, vocal or instrumental preference, and track length in your prompt.
Tip: For background music, first select instrumental and clearly specify the tempo and intended duration.
✓ Verified by official sourceModels: Lyria 3.5Source: Create your best tracks yet with Lyria 3.5 in Gemini.Shareable Bots and Verified badge in Marketplace
Grok Bot Marketplace has started listing addable Bots alongside plugins, with categories, search, and Bot pages. Bots created by the Grok Bot team carry a Verified badge.
Usage details
Who can use it: Grok Bot users; Bot addition and Enterprise access restrictions depend on administrator settings.
How to use it: Open the Bots category or search in Marketplace, then select Add from a Bot page.
Tip: The Verified badge helps verify the source, but review the connector permissions requested by each Bot individually.
✓ Verified by official sourceSource: Grok Bot ChangelogPlugin/skill marketplace search and shared settings center revamped
Work 3.2.5 improved search and display logic in the plugin and skill marketplace, and aligned Settings Center across Chat and Work modes.
Usage details
Who can use it: Users of Kimi Work and Chat modes.
How to use it: Use the Marketplace search field; open Settings Center for shared settings.
Tip: If you use the same skill in different modes, check that its permissions and connection settings are up to date in the shared center.
✓ Verified by official sourceSource: Kimi Work Release NotesIn-conversation search added
Work 3.2.5 introduced in-conversation search to quickly find records in long conversations.
Usage details
Who can use it: Kimi Work users.
How to use it: Open the search field in the conversation view and enter a keyword or phrase.
Tip: Search by source URL, file name, or task title to find previous output again.
✓ Verified by official sourceSource: Kimi Work Release Notesant apply manages resources as code
ant CLI 1.30.0 added the ant apply command, which creates and updates agents, environments, Skills, memory stores, and deployments from files. The claude-lock.json file written by the command ensures the same resources are updated in subsequent local or CI runs.
Usage details
Who can use it: ant CLI 1.30.0 and Claude Platform resources.
How to use it: Create resource definition files in the repository, run ant apply, and add the generated claude-lock.json file to version control.
Tip: Review the plan output before applying in CI, and use a separate lock file for production resources.
✓ Verified by official sourceSource: Claude Platform API sürüm notlarıGPT-6 Astra upgrades coding, research, and computer use
GPT-6 Astra was introduced to handle coding, research, computer use, and complex multistep tasks from start to finish; it can create documents, spreadsheets, and presentations based on a provided template. It was added to the Responses and Chat Completions endpoints in the API under the gpt-6-astra ID, and the Responses API is recommended for tool calling.
Usage details
Who can use it: Gradual rollout to ChatGPT Plus, Pro, Business, and Enterprise; available in the API, Microsoft Azure, and AWS Bedrock. Enterprise access may be off by default.
How to use it: Select GPT-6 Astra from the available model list in ChatGPT; use `gpt-6-astra` in the API and switch to the Responses API for tool calling.
Tip: Because Astra’s safety monitoring may stop a task, design pause and user review workflows for production agents.
✓ Verified by official sourceModels: GPT-6 Astra, gpt-6-astraSource: GPT-6 Astra duyurusuSource: ChatGPT yayın notlarıSource: OpenAI API changelogGrok Bot opens to Enterprise
Grok Bot is no longer limited to early-access teams and is now available to Enterprise customers. Access, network, and audit controls have been added; Grok and Cursor Enterprise customers received two weeks of free use and the ability to invite employees without seats.
Usage details
Who can use it: Grok and Cursor Enterprise customers; two weeks of free use from the announcement date, followed by enterprise plan terms.
How to use it: Enable Grok Bot in the admin dashboard and grant organization members access.
Tip: Configure access groups, network controls, and audit log policies before rolling it out widely.
✓ Verified by official sourceSource: Grok Bot for EnterpriseZendesk and OneNote plugins added in beta
The Zendesk plugin reviews support tickets and customer history, then drafts replies. The OneNote plugin finds and summarizes notes, extracts decisions and action items, and creates and updates notes using supported actions.
Usage details
Who can use it: In beta in the ChatGPT and Codex Plugin directory; users must connect their own Zendesk or OneNote accounts.
How to use it: Install Zendesk or OneNote from the Plugin directory and connect your personal account.
Tip: Limit the plugin connection to support records and notes the user already has access to.
✓ Verified by official sourceSource: ChatGPT yayın notlarıDingTalk Channel bots can send file attachments
DingTalk bots can now send the reports and documents they generate directly as file attachments instead of pasting them into the message. Users can open and download files from chats on their phones; bots can also include attachments in proactive notifications.
Usage details
Who can use it: Qwen Code Channel users connected to DingTalk.
How to use it: Set up the DingTalk Channel connection; the bot will automatically send an attachment when a file-generating task is complete.
Tip: Request reports as files when you plan to review them on mobile to preserve formatting and attachments.
✓ Verified by official sourceSource: Qwen Code Weekly: Pick How the AI Talks to You, DingTalk Bot Sends Files, Web Shell Gets a Built-in TerminalChanging effort mid-conversation enters beta on Google Cloud
Per-message effort changes are enabled in beta on Google Cloud for Claude Fable 5.1, Claude Mythos 5.1, and Claude Opus 5. Developers can change the output_config.effort value on subsequent turns without breaking the prompt cache.
Usage details
Who can use it: Applicable models on Google Cloud; requires the mid-conversation-output-config-2026-07-01 beta header.
How to use it: Add a role=system message and output_config.effort to the messages array.
Tip: Preserve the cache by using low effort for routine tasks and high effort on the final turn when verification is required.
✓ Verified by official sourceModels: Claude Fable 5.1, Claude Mythos 5.1, Claude Opus 5Source: Claude Platform API sürüm notlarıVietnamese, Indonesian, Urdu, Thai, and Bengali added to Grok Bot
Vietnamese, Indonesian, Urdu, Thai, and Bengali have been added to Grok Bot’s interface language options. The new Bot’s welcome message also follows the selected app language.
Usage details
Who can use it: In Grok Bot Language settings, on supported client versions.
How to use it: Select one of the supported languages under Settings → Language, or use Follow System.
Tip: When creating a new Bot for multilingual teams, check the default greeting language.
✓ Verified by official sourceSource: Grok Bot ChangelogLyria 3.5 music model reaches GA in the API
Lyria 3.5 was released as GA, with full-length songs, more consistent musical structure, natural vocals, and fine-grained control over duration and structure. It can generate 44.1 kHz stereo audio from text and image inputs.
Usage details
Who can use it: Gemini API and Google AI Studio; usage and pricing terms depend on the API plan.
How to use it: Set the model ID to lyria-3.5 in the Gemini API music generation guide.
Tip: Make editing easier by describing the song structure in the prompt using section names and approximate durations.
✓ Verified by official sourceModels: lyria-3.5Source: Release notes | Gemini API - Google AI for DevelopersQwen Code Web Shell gets a built-in terminal
A built-in terminal that runs alongside the chat has been added to the right panel of Web Shell. Each terminal tab is independent, so a long-running server can run in one tab while commands are executed in another.
Usage details
Who can use it: Qwen Code Web Shell users.
How to use it: Open a terminal tab from the chat’s right panel; commands run in the current workspace directory.
Tip: Keep the development server in one tab and run test and log commands in separate tabs.
✓ Verified by official sourceSource: Qwen Code Weekly: Pick How the AI Talks to You, DingTalk Bot Sends Files, Web Shell Gets a Built-in TerminalChoose and change Qwen Code’s response style on the fly
Qwen Code now offers four selectable built-in output styles: Concise, Proactive, Explanatory, and Learning. The style can be saved in settings.json, applied for a single session with --output-style, or changed mid-conversation with the /output-style command.
Usage details
Who can use it: Qwen Code v0.22.3–v0.23.0 CLI, Web Shell, and supported interfaces.
How to use it: Use the general.outputStyle setting, pass --output-style at startup, or run /output-style during a session.
Tip: Choose Learning/Explanatory when learning a codebase, and Concise when you only want the result.
✓ Verified by official sourceSource: Qwen Code Weekly: Pick How the AI Talks to You, DingTalk Bot Sends Files, Web Shell Gets a Built-in TerminalSites can be shared with specific people outside the workspace
Eligible Site owners can share a Site live with specific people outside the workspace without making it public. External users can view and use the Site, but cannot edit or publish it.
Usage details
Who can use it: Limited to eligible Site owners and workspace settings; external recipients must sign in with the authorized account.
How to use it: Open the Site, select Share, add the recipient’s email, and save viewer access.
Tip: Before sharing externally, check what data is visible through the Site’s connected apps.
✓ Verified by official sourceSource: ChatGPT yayın notlarıBots keep working on long tasks
Grok Bot has started continuing work on long jobs and page-by-page web browsing instead of stopping early. The latest version of a shared Bot and its integrations are retained when reopened.
Usage details
Who can use it: Grok Bot users; depends on task duration, plan, and computer resources.
How to use it: Give Bot a long task with a single goal and completion condition; let it run in the background.
Tip: Define interim outputs or checkpoints for long tasks; test fully open-ended tasks before making them routine.
✓ Verified by official sourceSource: Grok Bot ChangelogGemini 3.8 Flash reaches GA
Gemini 3.8 Flash was released as Google’s most capable Flash model for long-running software engineering, autonomous agents, and complex enterprise workflows. The model improves reasoning and coding over 3.7 Flash while retaining Flash-class speed and cost.
Usage details
Who can use it: Gemini API, Google AI Studio, and Gemini Enterprise; for consumers, Gemini, Search AI Mode, and Sheets for Google AI Pro and Ultra subscribers.
How to use it: Select the gemini-3.8-flash model in Gemini API or AI Studio.
Tip: For long-running agent tasks, measure costs per task to assess the additional token usage from higher effort levels.
✓ Verified by official sourceModels: Gemini 3.8 FlashSource: Release notes | Gemini API - Google AI for DevelopersSource: Introducing Gemini 3.8 Flash and 3.8 Flash Cybergrok-imagine-image-quality slug to be retired on November 2
xAI announced that the `grok-imagine-image-quality` model slug will be retired on November 2, 2026. After that date, requests will be served with `grok-imagine-image-2.0` and `quality: low`, while keeping the same request/response format; the new target has a lower price per image.
Usage details
Who can use it: API customers; migration will happen through automatic routing on November 2, 2026, but an explicit migration is recommended.
How to use it: Change the `model` field in your code to `grok-imagine-image-2.0` and explicitly manage quality with `low`, `medium`, or `auto`.
Tip: Instead of relying on automatic routing, update the model field in advance and test for quality and cost regressions.
✓ Verified by official sourceModels: grok-imagine-image-quality, grok-imagine-image-2.0Source: grok-imagine-image-quality Retirement on November 2, 2026Source: Release NotesMuse Spark 1.3 released for longer agent and coding tasks
Muse Spark 1.3 offers better plan tracking, collaboration with the user, clarification questions when there’s uncertainty, and management of multiple workflows during long tasks in multi-step agent and coding work. Meta reported approximately 20 percent fewer tool calls and 25 percent fewer tokens compared with Muse Spark 1.2.
Usage details
Who can use it: Available in Muse Code and Meta Model API; the Muse Spark 1.3 Standard and Contributor variants are available in the API, subject to usage terms.
How to use it: Select the model in Muse Code or use the muse-spark-1.3 model ID in a Meta Model API call; adjust the reasoning effort level for long tasks.
Tip: For ambiguous goals, have the agent summarize its plan first and ask for approval before irreversible steps.
Qwen3.8-Max-0902 snapshot released
Qwen3.8-Max-0902, an upgraded snapshot of qwen3.8-max, brings improvements to more complex engineering and long-horizon autonomous development, multi-tool agent coordination, and visual understanding of documents and charts. The 1M context, thinking mode, and tool ecosystem remain unchanged.
Usage details
Who can use it: Model Studio International API users; access and quotas depend on the account’s region.
How to use it: Use the qwen3.8-max-0902 or qwen3.8-max-2026-09-02 identifier when making Model Studio calls.
Tip: For long-running coding tasks, pin the snapshot identifier and A/B test results against the previous qwen3.8-max.
✓ Verified by official sourceModels: qwen3.8-max-0902, qwen3.8-max-2026-09-02Source: Model lifecycle and updatesAgentic video understanding is released
The Gemini API can now dynamically navigate a timeline and request transcripts, frames, or audio clips as needed, rather than processing long videos in their entirety in a static way. Google says this approach can use up to 88% fewer tokens than static processing for long content.
Usage details
Who can use it: Interactions API and GenerateContent API for Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite.
How to use it: Enable the agentic video configuration in the relevant request type, as described in the Gemini API video understanding guide.
Tip: For long videos, preserve your token budget by using the model’s selective navigation instead of splitting the entire video into frames in advance.
✓ Verified by official sourceModels: Gemini 3.7 Flash, Gemini 3.6 Flash, Gemini 3.5 Flash-LiteSource: Release notes | Gemini API - Google AI for DevelopersClaude Fable 5.1 and Claude Mythos 5.1 released
Claude Fable 5.1 is generally available for long-running agentic coding, knowledge work, and research; a less restricted version of the same model, Claude Mythos 5.1, is reserved for Project Glasswing and verified access programs. Both support a 1M-token context by default, a 128K maximum output, and continuous adaptive thinking.
Usage details
Who can use it: Fable 5.1 is available on Claude API, Amazon Bedrock, AWS, Google Cloud, and Microsoft Foundry; Mythos 5.1 is available through verified cybersecurity and life sciences programs, initially to a limited number of U.S. organizations.
How to use it: Use the claude-fable-5-1 model ID in the API; apply to the appropriate trusted access program for Mythos 5.1.
Tip: For preserved thinking and tool_choice compatibility in Fable 5.1, apply the request changes in the migration guide before upgrading.
✓ Verified by official sourceModels: Claude Fable 5.1, Claude Mythos 5.1Source: Claude Fable 5.1 ve Mythos 5.1 duyurusuSource: Claude Platform API sürüm notlarıHealthcare Public Data plugin for Clinicians
Eligible ChatGPT for Clinicians users in the U.S. can search nine public health data apps, including biomedical research, clinical trials, drugs, Medicare, and provider records. The apps are read-only and do not access patient charts; do not include protected health information in searches sent to public sources.
Usage details
Who can use it: Eligible ChatGPT for Clinicians users in the U.S.
How to use it: Install Healthcare Public Data from the Plugin directory and connect the apps you want to use.
Tip: Do not include PHI in searches of public sources; independently verify clinical decisions against primary medical sources.
✓ Verified by official sourceSource: ChatGPT yayın notlarıFable 5.1 and Mythos 5.1 cache read price drops to $0.25
Prompt cache reads for Claude Fable 5.1 and Claude Mythos 5.1 are priced at $0.25 per million tokens; this is 0.025 times the base input price and below the rate for previous models. Fable 5.1’s standard input and output prices remain the same as Fable 5.
Usage details
Who can use it: Claude API and supported cloud deployments with token-based pricing; Mythos access is limited to verified programs.
How to use it: Use prompt caching to resend the same context; specify the model ID as claude-fable-5-1 or, if authorized, claude-mythos-5-1.
Tip: For long, tool-heavy tasks, measure cache read savings by moving fixed instructions into the cache.
✓ Verified by official sourceModels: Claude Fable 5.1, Claude Mythos 5.1Source: Claude Fable 5.1 ve Mythos 5.1 duyurusuSource: Claude Platform API sürüm notlarıGuided vision comes to Gemini Live
Guided vision not only describes what the camera sees aloud but also gives users spoken guidance on reframing the shot, panning the camera, or centering an object. The feature was designed especially in collaboration with blind and low-vision communities.
Usage details
Who can use it: Rolling out gradually on phones running Android 9 and later, in countries where Gemini is supported; language and regional requirements apply.
How to use it: Share your camera in Gemini Live; enable Guided vision in the Gemini app settings, the TalkBack menu, or the Accessibility shortcut.
Tip: Use the feature for assistive tasks such as describing menus, labels, and objects, rather than for navigation or obstacle detection.
✓ Verified by official sourceModels: Gemini LiveSource: September Android Drop: Remember where you put things, ease motion sickness, and moreMuse Voice Transcribe real-time speech API launched
Meta has released Muse Voice Transcribe, a model from the Muse Spark family, with support for real-time streaming ASR, diarization for more than 20 speakers, endpointing, voice activity detection, and multilingual code-switching. Meta Model API offers WebSocket live streaming and a one-shot transcription endpoint for recorded files.
Usage details
Who can use it: Available in Meta Model API, Meta AI for Mac, and Muse Code; API pricing is $3 per 1,000 minutes, or about $0.18 per hour. Support is listed for 25 verified languages and additional languages.
How to use it: For live streaming, use wss://api.meta.ai/v1/asr/realtime; for recordings, use the POST /v1/asr/transcribe endpoint. Guide recognition with the keywords and languageBias parameters.
Tip: For meetings and calls, select DIARIZATION or ENDPOINTING mode and overwrite the previous text with cumulative partial transcripts.
✓ Verified by official sourceModels: Muse Voice TranscribeSource: Introducing Muse Voice TranscribeSource: Meet Muse Voice Transcribe, streaming speech to text on the Meta Model APIGemini remembers item locations in Find Hub
Users can tell Gemini where they put an item without a tracking tag, such as a passport or spare key, and optionally add a photo. Later, they can ask Gemini or find the location in the remembered items tab in Find Hub.
Usage details
Who can use it: On devices running Android 16 and later, in countries where Gemini and Find Hub are supported.
How to use it: Give a command such as ‘Hey Google, remember that I put my passport in the bedroom drawer in Find Hub’; then ask Gemini where it is.
Tip: To keep remembered locations up to date, edit or delete the old entry when an item is moved.
✓ Verified by official sourceModels: GeminiSource: September Android Drop: Remember where you put things, ease motion sickness, and moreBiosafety and refusal performance announced for Grok 4.6
xAI announced Grok 4.6’s performance in refusing concealed biosafety hazards while continuing routine biology tasks in LatchBio’s independent BioSecBench evaluation. This is not a new model release, but the publication of measurements on the model’s safety tuning and biosafety capabilities.
Usage details
Who can use it: On existing API and product surfaces that use Grok 4.6; the announcement does not specify a new access region or plan.
How to use it: No separate setting is required; the safety behavior is built into the deployment of the Grok 4.6 model.
Tip: Use model outputs in biology workflows alongside expert review and your organization’s biosafety procedures.
Pull request previews and new Bot in sections
Pull request links in Grok Bot messages now show the number, title, and open/draft/merged/closed status. Users can also create a new Bot directly from a section in the sidebar.
Usage details
Who can use it: Grok Bot desktop app users.
How to use it: Open a pull request link in a Bot chat; create the new Bot from the section you want in the sidebar.
Tip: When requesting a code review, also verify the PR status and target branch; the preview shows only link metadata.
✓ Verified by official sourceSource: Grok Bot Changelog
No updates match these filters.