YZ·RADARyz.mikail.net
Daily edition · 11 November 2026

Gemini 4 Carbon is generally available; Meta AI is coming to Turkish on AI glasses

The TOP 10 is unchanged from yesterday. In Artificial Analysis Intelligence Index v5.0.1, Claude Fable 5.2 leads with 65 points, Gemini 4 Carbon is second with 60, and Claude Opus 5.5 is third with 59; neither rankings nor scores changed among the top ten. The watchlist is also unchanged: Muse Wonder 44, Claude Haiku 5.5 and Muse Ember 36, Grok 4.7 and Tencent Hy4 35, Qwen3.8 Max 34. The latest accessible view of the LMArena text leaderboard is still dated October 27; Gemini 4 Carbon and Gemini 4 Argon are in the top two spots there, while Fable 5.2 has not appeared yet. Although the rankings remain static, the day's biggest news came from the model in second place: Gemini 4 Carbon is no longer available only to select organizations.

Google made Gemini 4 Carbon generally available on November 10. Developers can use the model in Gemini API and Google AI Studio with the `gemini-4-carbon` identifier, but only on the paid tier: through November 30, $2 per 1M input tokens and $10 per 1M output tokens; starting December 1, $4 / $20. Google AI Ultra subscribers are getting access to the model gradually in the Gemini app, Google Antigravity, and AI Mode in Search. It is also GA in Gemini Enterprise across web, mobile, Workflow Builder, and developer tools; it is off by default, so an administrator must enable it. BigQuery’s generative AI functions also support Carbon. The defensive cybersecurity version, Gemini 4 Carbon for cyber, is GA, but remains behind the Fairwind Program allowlist. The November Pixel Drop, also released that day, includes Create My Widget, which builds a personal home-screen widget from natural-language prompts; Gemini Call Screen, which screens calls by talking to the caller on Pixel 11 (US); and task automation expanding to the UK, India, and Japan. Skills is starting to roll out across Gmail, Docs, Slides, Drive, and Chat in Workspace (English only for now). Gemini summaries and Help me write in Gmail are rolling out to all Workspace users, and Workspace Studio flows have gained a `Wait until` step that can wait for up to 14 days. The Gemini CLI v0.69 preview displays daily token usage as a heat map with the `/stats heatmap` command.

Two dates matter for OpenAI. Advanced and Standard Voice in ChatGPT are being discontinued on December 11; voice mode will move to GPT-Live after that date, while live video and screen sharing will remain in Advanced Voice for now. `gpt-image-2` and `gpt-image-2-2026-04-21` have been deprecated in the API and will be removed on February 10, 2027; the recommended replacements are `gpt-image-2.6-sunburst` and `gpt-image-2.6-flare`. Businesses in ChatGPT has launched in beta: businesses can claim a profile and update their product feed in Business Hub, and eligible businesses in the US can publish their own Business Agent. The Python, Node, and Ruby SDKs now support typed async function tools. Codex 0.169 alpha releases add model and reasoning effort selection in Cloud conversations, thread archiving and restoration, and device-code sign-in for headless environments such as SSH. Anthropic has clarified Fable 5.2’s migration rules: thinking is always adaptive, requests that send `budget_tokens` receive a 400 error, only `auto` and `none` are supported for `tool_choice`, and thinking blocks are valid only for the account that generated them. The model is available on all paid Claude plans; in Enterprise, an administrator must enable it. Claude Code 2.1.321 and the desktop gateway have logged 33 fixes related to memory in long sessions, TLS disconnections, and hook behavior. There is news from Meta that directly affects Türkiye: starting the week of November 11, Meta AI will be available on screenless AI glasses such as Ray-Ban Meta, Oakley Meta, and Meta Glasses in 12 new countries, including Türkiye, and in 10 new languages, including Turkish. Muse has also arrived on Apple Watch and Wear OS; Instagram’s Effect Builder, which creates Stories effects from prompts, has launched in the US.

Qwen Code 0.27.2 stands out at Alibaba. WeChat and QQ channels can be connected by QR code, along with an email channel, from Channel Manager in Web Shell. Mods extensions can make one-off model calls with `$.model.complete` and store data across sessions with `$.store`; Managed Agent has gained dedicated `/btw` follow-ups and session recaps, and pending approvals now trigger terminal and desktop notifications. HiRoute 0.6.0 adds reasoning effort selection on Codex fixed plans. At Moonshot, with Kimi Code 2.5.3 (November 10), a steer message sent to a running agent now interrupts the active model request and continues in the same turn; there is also a new `POST /api/v1/sessions/{session_id}/steer` endpoint for this. Stdio MCP servers defined with scripts such as `npx` now also start on Windows. The main branch has 19 more changes, including error codes for the local server API, creating sessions with `agent_config`, and tracking background tasks with `task_id`; these will arrive in the next release. We have also added a development we missed in previous days to the archive: Kimi K3.1 Pro Highspeed has been available on Amazon Bedrock since November 6, running in the Flex tier at half the Standard price. There is nothing new on DeepSeek’s own channels; a Microsoft Foundry post updated on November 9 lists an Azure deployment for DeepSeek-V4.1-Flash-1017, but this has not been confirmed by DeepSeek. Tencent has only a GroupAssistBench data update; there have been no new changes at xAI, Z.ai, or Xiaomi since November 9. A total of 164 new developments were added to the archive today.

Today’s TOP 10

The ten best unique models by intelligence index; each model is counted at its highest-scoring setting.

  1. 01

    Claude Fable 5.2

    Anthropic · Claude=
    65pts

    Released on November 9 and leads with 65 points. Available on all paid Claude plans (an administrator must enable it in Enterprise). In the API (`claude-fable-5-2`, $8 / $40), thinking is always adaptive; requests that send `budget_tokens` receive a 400 error.Released: 9 November 2026

  2. 02

    Gemini 4 Carbon

    Google · Gemini=
    60pts

    Made generally available on November 10: `gemini-4-carbon` in Gemini API and AI Studio (paid tier only; $2 / $10 through November 30, $4 / $20 starting December 1). Gradually rolling out to Ultra subscribers in the Gemini app, Antigravity, and AI Mode.Released: 28 October 2026

  3. 03

    Claude Opus 5.5

    Anthropic · Claude=
    59pts

    Anthropic’s flagship Opus model; available on Claude Pro/Max/Team/Enterprise and in the API. Claude Code 2.1.322 added fast mode support on Vertex (global, US, EU).Released: 22 September 2026

  4. 04

    GPT-6 Astra

    OpenAI · ChatGPT=
    57pts

    OpenAI’s most powerful GPT-6 model. In Agents SDK Python v0.24.0-alpha.7, the default reasoning effort for this model increased from `low` to `medium`.Released: 3 September 2026

  5. 05

    GPT-6.1 Sol

    OpenAI · ChatGPT=
    55pts

    The default model on paid ChatGPT plans; API price is $2 / $10. The default reasoning effort became `medium` in an Agents SDK alpha release.Released: 29 September 2026

  6. 06

    Claude Sonnet 5.5

    Anthropic · Claude=
    55pts

    The most powerful model also available on the claude.ai free plan; tied with GPT-6.1 Sol at 55 points.Released: 28 September 2026

  7. 07

    GPT-6.1 Terra

    OpenAI · ChatGPT=
    51pts

    Released on October 26 in ChatGPT Work, Codex, and the API; focused on images and computer use. The admin preview of Dots agents runs through November 18.Released: 26 October 2026

  8. 08

    Gemini 4 Argon

    Google · Gemini=
    50pts

    Available on Google AI Pro/Ultra and the paid API; the $2 / $10 introductory price runs through November 30. Long decode output of up to 1 million tokens in a single request became GA in the Interactions API.Released: 30 September 2026

  9. 09

    Muse Wonder Full

    Meta · Meta AI=
    50pts

    Meta’s most powerful Muse model; no public API. The Muse app is now also available on Apple Watch and Wear OS.Released: 12 October 2026

  10. 10

    Kimi K3.1 Pro

    Moonshot AI · Kimi=
    46pts

    API at $2 / $10; open weights on Hugging Face (2.8T total / 104B active, 1M context). The Highspeed version has been on Amazon Bedrock since November 6 (`global.moonshotai.kimi-k3.1`, half price in the Flex tier). In Kimi Code 2.5.3, a steer message directs a running agent in the same turn.Released: 23 October 2026

Key signals

  • The TOP 10 is unchanged: Claude Fable 5.2 leads with 65 points, Gemini 4 Carbon has 60, and Claude Opus 5.5 has 59; the watchlist is also unchanged from yesterday.
  • Google: Gemini 4 Carbon is generally available. In the API, `gemini-4-carbon` costs $2 / $10 through November 30 ($4 / $20 starting December 1); access is rolling out to Ultra subscribers in the Gemini app and AI Mode.
  • Meta: Meta AI will be available on AI glasses in Türkiye and in Turkish starting the week of November 11; Muse has arrived on Apple Watch and Wear OS.
  • OpenAI: Advanced and Standard Voice will give way to GPT-Live on December 11; `gpt-image-2` will be removed from the API on February 10, 2027.
  • Anthropic: In Fable 5.2, thinking is always adaptive and `budget_tokens` triggers a 400 error; the model is available on all paid Claude plans.
  • 164 new developments: Alibaba 37, Anthropic 37, OpenAI 34, Google 27, Moonshot 22, Meta 5, DeepSeek 1, Tencent 1. No new changes at xAI, Z.ai, or Xiaomi.

Worth trying today

01

Try Gemini 4 Carbon in your own work before the introductory pricing ends

In Google AI Studio, select a project with billing enabled (Carbon is not available on the free tier), create an API key, and set the model to `gemini-4-carbon`. The price is $2 / $10 through November 30 and doubles to $4 / $20 starting December 1. Run some of your most challenging tasks—for example, a code change involving multiple files, a long document analysis, or a multi-step agent task—in both Carbon and the model you currently use; compare quality and total cost at the December 1 price. Remember that thinking tokens are also billed. If you are an Ultra subscriber, check whether Carbon appears in the model selector in the Gemini app; the rollout is gradual.

Google · Gemini
02

Get ready for Turkish Meta AI if you own Ray-Ban Meta glasses

Update the Meta AI mobile app and your glasses’ software to the latest version (new language and country support is listed in the v30 release notes). The rollout starts the week of November 11 and is gradual; if the feature does not appear right away, check again in a few days. Once it is available, start with short, direct commands, and note in your first few tries whether responses are in Turkish and which features are enabled in Türkiye.

Meta · Meta AI
03

Start migrating workflows that use `gpt-image-2` and the legacy voice mode

Find instances of `gpt-image-2` or `gpt-image-2-2026-04-21` in your code (for example, `grep -rn "gpt-image-2\b" .`) and replace them with `gpt-image-2.6-sunburst` or `gpt-image-2.6-flare`, then rerun your image-generation and editing tests; the deadline is February 10, 2027. If you use Advanced or Standard Voice in ChatGPT, try GPT-Live now from Settings > Voice, since the transition will be automatic after December 11. Live video and screen sharing will remain in Advanced Voice for now; keep an eye on this if you have a workflow that depends on those features.

OpenAI · ChatGPT
04

Clean up thinking and tool_choice settings in API code migrating to Fable 5.2

If your requests include `thinking: {type: "enabled", budget_tokens: ...}`, remove it or change it to `{type: "adaptive"}`; use effort to control the depth of thinking. Set `tool_choice` to `auto` or `none`, since `any` and forcing a specific tool are not supported. If you transfer conversation history between accounts, do not carry over thinking blocks; they are valid only for the account that generated them. If you use Claude Enterprise, an administrator must enable Fable 5.2 in the model access section of the organization settings.

Anthropic · Claude

This edition’s updates by AI

What changed in this edition for every AI in the TOP 10 and on the watchlist.

Anthropic · Claude

#1 Claude Fable 5.2#3 Claude Opus 5.5#6 Claude Sonnet 5.5Claude Haiku 5.5
Full history (836) →

New in this edition · 37

  • APIClaude API

    Fable 5.2 thinking blocks can only be used by the account that generated them

    Thinking blocks generated by Fable 5.2 and Mythos 5.2 work only with the account that generated them or a linked account. If another account submits one, the API drops the block; with the beta header, this may appear in input_transformations as organization_binding_mismatch.

    Usage details

    Who can use it: Claude API and Google Cloud Google-managed model access; integrations that share transcripts across accounts.

    How to use it: If you transfer conversations across accounts, don't share thinking blocks, or handle the input_transformations field on return using the thinking-binding-controls-2026-08-01 beta header.

    Tip: When storing transcripts, save the model, account, and organization IDs alongside the thinking blocks.

    ✓ Verified by official sourceModels: Claude Fable 5.2, Claude Mythos 5.2Source: Claude Fable 5.2 ve Claude Mythos 5.2 migration guideSource: Claude Fable 5.2 yenilikleri
  • AvailabilityClaude

    Fable 5.2 is now available on all paid plans; Enterprise admins must enable it

    Fable models are now available on the Pro, Max, Ultra, Team, and Enterprise paid plans; the Free plan is excluded. On Enterprise, the model remains disabled until an admin enables Fable 5.2. Seven days after launch, Fable’s active role and defaults will be migrated to Fable 5.2.

    Usage details

    Who can use it: Claude web, Mobile, Desktop, Cowork, Code, Design, Microsoft 365, and Claude Tag; paid plans. Enterprise requires model access to be granted by an organization admin.

    How to use it: Select Fable 5.2 from the model picker in Web, Desktop, or Mobile. An Enterprise admin must enable Fable 5.2 under model access in Organization settings.

    Tip: Check in advance whether Fable usage for Pro and Team standard seat users may draw from usage credits.

  • APIClaude API

    Fable 5.2 and Mythos 5.2 require adaptive thinking

    Thinking always runs adaptively in these models; sending budget_tokens with thinking.type set to disabled or enabled returns a 400 error. Forcing tool_choice to any or to a specific tool is also unsupported; use auto or none.

    Usage details

    Who can use it: Claude API and compatible platforms that offer these models; API integrations.

    How to use it: Remove the thinking field from the request or send {type: "adaptive"}; control depth with effort. Use auto or none for tool_choice.

    Tip: Before going to production, test legacy manual thinking and forced tool-call flows separately with the new model.

    ✓ Verified by official sourceModels: Claude Fable 5.2, Claude Mythos 5.2Source: Claude Fable 5.2 ve Claude Mythos 5.2 migration guideSource: Claude Fable 5.2 yenilikleri
  • CapabilityClaude Code

    If memory runs out while reading a transcript, new messages and SDK sessions are preserved

    A bug that caused new messages to be lost and claude -p, SDK, or VS Code sessions to hang when memory ran out while reading a transcript has been fixed. The session now handles the error more reliably.

    Usage details

    Who can use it: Claude Code 2.1.321 and later; CLI, SDK, and VS Code sessions using long transcripts.

    How to use it: Update the version; continue in the affected session while keeping a transcript backup.

    Tip: Regularly splitting very long sessions with a new summary or a separate session reduces memory pressure.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • APIClaude Code

    AskUserQuestion responses can no longer set the hidden afkTimeoutMs field

    In claude -p and SDK sessions, attempts by a host to change the timeout by adding the undocumented afkTimeoutMs field to a permission response are now rejected. Other AskUserQuestion flows remain unchanged.

    Usage details

    Who can use it: Claude Code CLI and Agent SDK; AskUserQuestion permission responses provided by a host application.

    How to use it: Remove the afkTimeoutMs field from your host integration; continue using supported AskUserQuestion fields.

    Tip: Validate SDK schemas during upgrades and avoid depending on undocumented fields.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • SettingClaude Code

    Auto default now saves the correct permission selection in /config

    When a new session starts in Auto mode, /config no longer shows Manual. A selection made with Space on Auto (default) is saved as Don't Ask; the previous incorrect Plan setting is no longer saved.

    Usage details

    Who can use it: Claude Code 2.1.321 and later; terminal sessions.

    How to use it: Update Claude Code to 2.1.321 or later; open /config and select a permission mode.

    Tip: Before changing the session's starting permission policy, check that your selection in /config was actually saved.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • APIClaude Code

    Failed NotebookEdit no longer reports success to hooks or telemetry

    If a notebook changes while the permission prompt is open and the edit fails, NotebookEdit no longer sends a success signal to hooks or telemetry.

    Usage details

    Who can use it: Claude Code 2.1.321 and later; hooks and OpenTelemetry pipelines using NotebookEdit.

    How to use it: Update the version; evaluate the NotebookEdit result in the hook along with its success code.

    Tip: In automations, check the success field of the edit result, not just whether the tool call completed.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • APIClaude Code

    Ask reasons and output matching fixed for multiple PreToolUse hooks

    In applications embedding Claude Code, one hook's ask reason no longer gets mixed up with another hook's reason; the previous ask response is also no longer lost when defer completes. Fixed an issue where a large hook output pointed to the wrong hook's file.

    Usage details

    Who can use it: Claude Code 2.1.321 and later; embedded applications and PreToolUse hooks.

    How to use it: Upgrade to 2.1.321 without changing your hooks; retest results if multiple hooks inspect the same tool call.

    Tip: Logging the hook name and call ID together makes debugging easier.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • CapabilityClaude Code

    Memory usage is lower in long sessions that read single lines from large files

    Fixed high memory usage caused by repeatedly reading single lines from large files in long sessions. This reduces the risk of long-running sessions slowing down or closing due to memory pressure.

    Usage details

    Who can use it: Claude Code 2.1.321 and later; long-running terminal, SDK, and desktop sessions.

    How to use it: Install the latest version; no need to migrate existing sessions separately.

    Tip: For very large files, request the necessary range in a single Read call instead of reading single lines in a loop.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • APIClaude Code

    The claude agents command now correctly interprets flags written before it

    Options written before claude agents—--permission-mode, --dangerously-skip-permissions, --model, --effort, or --agent—are no longer ignored when used with another flag such as --verbose.

    Usage details

    Who can use it: Claude Code CLI 2.1.321 and later; command-line and script usage.

    How to use it: Run claude --model <model> agents or a similar existing command sequence with 2.1.321 or later.

    Tip: In CI scripts, use one consistent format rather than mixing options before and after the subcommand.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • CapabilityClaude Code

    In Cloud sessions, pkill, pgrep, and pidwait no longer match Claude's main processes

    In Cloud sessions, pkill, pgrep, and pidwait no longer accidentally match the session's own parent processes. This reduces the risk of commands targeting Claude's own runtime infrastructure matching the wrong processes.

    Usage details

    Who can use it: Claude Code Cloud sessions; 2.1.321 and later.

    How to use it: Start the Cloud session with version 2.1.321 or later; no additional settings are needed for the commands.

    Tip: Even so, use a narrow name or PID filter in bulk process cleanup commands.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • CapabilityClaude API

    High-risk cyber and biology prompts for Fable 5.2 can be routed to Opus

    When Fable 5.2's safety classifiers block high-risk cyber or biology requests, Claude Opus 4.8 or Claude Opus 5 can be offered as a suitable fallback. This separates the safety block from the continuation of technical work.

    Usage details

    Who can use it: Claude Fable 5.2; flows where the relevant safety classifiers are applied in Claude API, Claude Code, Claude.ai, and Cowork.

    How to use it: In the API, handle refusal and fallback behavior using the stop_reason/refusal fields; use fallbacks: "default" for supported server-side fallback.

    Tip: Label fallback output separately from the primary model's output and log the safety category.

  • Pricing / PlanClaude Desktop

    Desktop spend limit warnings no longer disappear when the Gateway policy has no code rule

    Fixed an issue that prevented Claude Desktop from showing spend limit warnings when the Claude apps gateway policy contained no code rule at all. Spend limit notifications are visible again behind a compatible Gateway.

    Usage details

    Who can use it: Organizations using Claude Desktop and Claude apps gateway; spend limit warnings.

    How to use it: Upgrade Gateway and Desktop to 2.1.321 or later; do not change the existing spend limit setting.

    Tip: Test whether users see limit warnings with a real limit scenario behind Gateway.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • APIClaude apps gateway

    Gateway warns about new desktop settings instead of rejecting them outright

    When a desktop setting is newer than the version supported by Gateway, such as the oauth.omitResource field in a managed MCP server, Gateway no longer refuses to start. It warns about the unknown new field and continues serving.

    Usage details

    Who can use it: Claude apps gateway 2.1.321 and later; organizations using new Claude Desktop managed settings.

    How to use it: Update Gateway to 2.1.321; review the log warning and plan to move to a Gateway version that supports the field.

    Tip: When using a new field in managed settings, upgrade the Gateway and Desktop versions together.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • CapabilityClaude Code

    When the first model is blocked before responding, the fallback model no longer gives an incorrect refusal

    Fixed an issue where the fallback model would sometimes refuse the request or refer to an interrupted response when the first model was blocked before responding. The fallback now handles the new request with the correct context.

    Usage details

    Who can use it: Claude Code 2.1.321 and later; sessions using model fallback.

    How to use it: Use an up-to-date version of Claude Code; no fallback configuration or model selection changes are needed.

    Tip: In fallback tests, separately measure the scenario where the first model is blocked before responding.

    ✓ Verified by official sourceModels: Claude Fable 5.2Source: Claude Code v2.1.321 release notes
  • AvailabilityClaude Code

    If macOS MDM policy loads slowly, it is retried once at startup

    If a valid MDM profile loads slowly, Claude Code now retries managed policy settings once during macOS startup. Timeouts are also reported explicitly as timeouts instead of general read errors.

    Usage details

    Who can use it: Claude Code macOS; devices managed with MDM.

    How to use it: Update Claude Code; allow the MDM profile to be read on the device and restart the session.

    Tip: For enterprise deployments, schedule MDM profile application before Claude Code starts.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • Plugin / IntegrationClaude Code

    Old settings and secrets from managed marketplace takeovers can be restored

    When a managed extraKnownMarketplaces source takes over another source, old entries are set aside for the duration of cleanupPeriodDays. Settings and secrets can be restored with claude plugin restore-set-aside.

    Usage details

    Who can use it: Organizations using the Claude apps gateway or managed marketplace policies; Claude Code 2.1.321 and later.

    How to use it: After upgrading, review the entries that were taken over; use the claude plugin restore-set-aside command if needed.

    Tip: Before migrating marketplaces, export existing plugin options and secrets to create a rollback plan.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • Plugin / IntegrationClaude Code

    Claude in Chrome processes no longer linger after closing behind a proxy

    Fixed an issue where Claude in Chrome and computer use server processes continued running after the parent process closed when a proxy connection stalled. Orphaned browser processes are now cleaned up more reliably when a session ends.

    Usage details

    Who can use it: Claude Code, Claude in Chrome, and computer use; sessions behind a proxy.

    How to use it: Upgrade Claude Code to 2.1.321 or later; you may need to manually terminate existing orphaned processes once.

    Tip: If you use long keep-alive periods on a proxy, monitor session shutdown logs.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • AgentClaude Code

    Pending background agent reports after Stop no longer overload a single request

    Fixed an issue where too many pending background agent reports were sent in a single request after Claude was stopped. The main conversation now receives a more controlled amount of report content.

    Usage details

    Who can use it: Claude Code 2.1.321 and later; terminal and desktop sessions using background agents.

    How to use it: Use the latest version; retest workflows that stop and restart many agents at once.

    Tip: Splitting agent tasks into smaller groups also reduces context usage in the main session.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • APIClaude Code

    Prompt suggestions no longer make extra API calls for responses ending in a tool call

    Fixed an issue where prompt suggestions sent unnecessary additional API requests when the model responded with a tool call. This reduces unnecessary latency and usage.

    Usage details

    Who can use it: Claude Code 2.1.321 and later; sessions with prompt suggestions enabled.

    How to use it: Update the version; no additional setting or beta header is required.

    Tip: Compare the number of requests generated by prompt suggestions before and after updating when tracking API costs.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • CapabilityClaude Code

    Windows watchPaths hooks no longer stop on lowercase drive paths

    A bug that caused FileChanged hooks to stop after the first shell command when the working directory’s drive letter was lowercase on Windows has been fixed. File changes monitored with watchPaths now trigger throughout the session.

    Usage details

    Who can use it: Claude Code Windows sessions; projects using watchPaths.

    How to use it: Update Claude Code to 2.1.321 or later; continue using watchPaths without changing your settings.

    Tip: Using a consistent drive-letter format in Windows configurations can still help reduce path-comparison issues.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • CapabilityClaude Code

    If TLS disconnects mid-response, the partial response is preserved or the request is retried

    When a connection drops during a response and causes an EPROTO TLS error, Claude Code now preserves the partial response or retries the request. This restores the resilient behavior of previous versions.

    Usage details

    Who can use it: Claude Code 2.1.321 and later; API, proxy, or gateway environments with unstable connections.

    How to use it: Update the version; in environments that use a proxy, restart the regular Claude Code session.

    Tip: For critical, long-running tasks, keep transcript backups enabled so you don’t lose partial response text.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • AvailabilityClaude Code

    Unresponsive TLS networks no longer leave sign-in and token operations waiting indefinitely

    Fixed an issue where sign-in, token refresh, and logout operations could wait indefinitely if a network connection was accepted but no TLS handshake response arrived.

    Usage details

    Who can use it: Claude Code 2.1.321 and later; environments using a proxy, VPN, or TLS inspection.

    How to use it: Install the latest version; if the connection issue persists, check the proxy’s TLS handshake response and timeout settings.

    Tip: Wrap authentication commands in a timeout in CI so network failures do not leave jobs running indefinitely.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • AgentClaude Code

    Subagent progress summaries no longer run unnecessarily to completion when the model returns a tool call

    Subagent progress summaries no longer keep running unnecessarily until completion after the model responds with a tool call. This reduces wait time and unnecessary model calls.

    Usage details

    Who can use it: Claude Code 2.1.321 and later; background subagent progress summaries.

    How to use it: Update the version; start subagents as usual.

    Tip: When running many subagents, monitor progress notifications’ call counts in telemetry.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • FeatureClaude Code

    Inter and Space Grotesk added to the recommended defaults in the Artifact design guide

    Artifact design guidelines now consider the Inter and Space Grotesk fonts suitable default options. This affects Claude’s typography choices when generating new Artifacts.

    Usage details

    Who can use it: Claude Artifact generation; updated design guide with Claude Code 2.1.321 and later.

    How to use it: When generating an Artifact, use the default design guidance unless you want a specific font; if you need a particular brand font, state it explicitly in the prompt.

    Tip: For accessibility, request a readable body font instead of a decorative font.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • FeatureClaude Code

    Guidance on signing in and copying Artifacts from another organization is clearer

    When an Artifact belongs to another organization, Claude now recommends /login only if the Artifact is in one of your organizations. It also explains more clearly where to find a copy after an account transfer.

    Usage details

    Who can use it: Multi-organization users working with Claude Code and Artifact; 2.1.321 and later.

    How to use it: Check the organization information in the Artifact message; sign in to the correct organization account, then copy the Artifact if needed.

    Tip: Export important Artifact links and copies before switching organizations.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • CapabilityClaude Code

    Skill and command names containing Bamum characters that appear blank in some fonts are rejected

    Skill and command names containing a Bamum letter that some fonts render as blank are no longer accepted. This prevents the use of names that are invisible or indistinguishable in menus.

    Usage details

    Who can use it: Claude Code 2.1.321 and later; skill and slash command authors.

    How to use it: Rename the skill or command using Latin letters and safe separators, then reload the plugin.

    Tip: Limit names in shared plugins to visible Unicode characters.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • FeatureClaude Code

    Sorting guidance for large Artifact databases now recommends reading page by page

    When an Artifact database collection is too large to sort, the Artifact tool no longer suggests adding filters. Claude is instructed to read and sort the collection page by page.

    Usage details

    Who can use it: Claude Code and the Artifact tool; sessions using large database collections.

    How to use it: Use the Artifact database in the latest version of Claude Code; follow Claude’s pagination guidance for large collections.

    Tip: For large collections, limit the page size and required fields before sorting.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • CapabilityClaude Code

    claude -p now reports the duration of background commands accurately

    claude -p no longer incorrectly reports that long-running background commands stopped after 10 minutes took 30 minutes. The duration reported to Claude now matches the actual behavior.

    Usage details

    Who can use it: Claude Code 2.1.321 and later; claude -p and automation runs.

    How to use it: Update Claude Code; no additional setting needs to be enabled or disabled.

    Tip: In automation prompts, use the actual status reported by the command result rather than assuming a fixed duration.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • FeatureClaude Code

    The dataviz skill recommends a separate meter row for each ratio

    The bundled dataviz skill now recommends a separate meter row for each ratio when multiple ratios have their own limits. This makes it possible to see each limit separately instead of using a single overall meter.

    Usage details

    Who can use it: Claude Code 2.1.321 and later; chart generation using the bundled dataviz skill.

    How to use it: Specify the ratios and limits in your dataviz request; the updated skill will recommend a separate row for each one.

    Tip: In dashboard designs, include the ratio's unit and limit in the meter heading.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • CapabilityClaude Code

    The command panel and prompt reappear after a dialog closes

    Fixed an issue where the command panel disappeared after a dialog closed and the prompt remained hidden until a redraw. The panel and input field now return as expected.

    Usage details

    Who can use it: Claude Code terminal interface 2.1.321 and later.

    How to use it: Update Claude Code; if the issue occurs, reopen with the latest version rather than waiting for a redraw in the same session.

    Tip: If you see a temporary blank screen in the terminal, check the panel state first instead of clearing the entire session with Ctrl+L.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • CapabilityClaude Code

    Install scripts and trusted TLS certificate loading are faster

    claude install and install scripts now complete about 2-4 seconds faster. Trusted TLS certificates also load faster at startup.

    Usage details

    Who can use it: Claude Code 2.1.321 and later; installation and every session startup.

    How to use it: Rerun claude install or the existing install script with the latest version.

    Tip: When comparing enterprise CI times, measure first runs and cached runs separately.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • APIClaude Code

    Unnecessary SEP-2352 warning removed when logging in to MCP again

    When logging in again to an MCP server saved by Claude Code, claude mcp login no longer prints a SEP-2352 issuer warning every time.

    Usage details

    Who can use it: Claude Code 2.1.321 and later; users logging in again to saved MCP servers.

    How to use it: Run claude mcp login <server> as usual; no additional setting is required.

    Tip: If you still see a warning after logging in again to an OAuth server, check the version first instead of deleting and re-adding the server.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • APIClaude Code

    Non-interactive telemetry hook counts no longer double-count built-in hooks

    After Bash and PowerShell calls, two built-in hooks are no longer added to the num_hooks and num_success fields in OpenTelemetry hook events for non-interactive sessions.

    Usage details

    Who can use it: Claude Code non-interactive sessions; OpenTelemetry hook metrics.

    How to use it: Update to 2.1.321; use the new counter baseline in your telemetry consumer.

    Tip: Add a version breakdown when comparing counters from old and new versions on the same chart.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • APIClaude Code

    OpenTelemetry vcs fields are no longer missing in previously launched Desktop processes

    OpenTelemetry repository attributes are no longer missing from Claude Desktop sessions opened from a previously launched process. vcs.* fields are now correctly included in the telemetry pipeline.

    Usage details

    Who can use it: Enterprise monitoring setups using the Claude Desktop Code tab and OpenTelemetry.

    How to use it: Update Claude Code; open a new session without reconfiguring your exporter and check the vcs.* fields.

    Tip: In dashboards, read repository identity from vcs.* fields as well as process-start labels.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • SettingClaude Code

    Remote Control’s initial prompt accepts a y response without a trailing newline

    If terminal input in the initial claude remote-control prompt ends with y and does not include Enter, Claude Code no longer silently ignores it. The response is treated as confirmation.

    Usage details

    Who can use it: Claude Code 2.1.321 and later; Remote Control launched with scripts, pipes, and batch jobs.

    How to use it: Update the claude remote-control command; if you send y to the initial prompt through a pipe, adding a separate Enter is no longer required.

    Tip: For automation, printf 'y\n' is still more portable across different terminal environments.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes
  • Plugin / IntegrationClaude Code

    Built-in plugin type files are no longer written accidentally to the builtin folder

    Fixed an issue where type files generated for Claude Code's built-in plugins created an unwanted builtin folder in the project when using --plugin-dir . or --plugin-dir ./builtin.

    Usage details

    Who can use it: Claude Code 2.1.321 and later; plugin developers using --plugin-dir.

    How to use it: Update the version and restart the plugin with claude --plugin-dir <folder>.

    Tip: Before adding generated files to version control, check the working tree for an unexpected builtin folder.

    ✓ Verified by official sourceSource: Claude Code v2.1.321 release notes

Google · Gemini

#2 Gemini 4 Carbon#8 Gemini 4 Argon
Full history (244) →

New in this edition · 27

  • Plugin / IntegrationAgent Platform Workbench / Antigravity CLI

    Antigravity CLI integrated into Agent Platform Workbench instances

    Antigravity CLI is now available in preview on Agent Platform Workbench instances that use the Debian 12 (`workbench-instances-2603`) image or later. The CLI supports creating, describing, editing, validating, and running notebooks; creating Jupyter kernels; `gcloud`/`bq` commands; and the `%%agy` cell magic.

    Usage details

    Who can use it: Agent Platform Workbench images running Debian 12 or later; enabled by default on new instances, in preview.

    How to use it: Open JupyterLab in a Workbench instance, go to File > New launcher > Antigravity CLI tile, or run `agy` in a terminal; approve the suggested file/command operations.

    Tip: Because the CLI writes directly to the notebook on disk, save the notebook before making changes and review generated code before running it.

  • ModelGemini 4 Carbon

    Gemini 4 Carbon is generally available in the API and AI Studio

    Gemini 4 Carbon has reached GA for complex software, science, multi-step agent, and cybersecurity tasks. The model is available through the Gemini API/Google AI Studio, Gemini Enterprise Agent Platform, and enterprise developer workflows.

    Usage details

    Who can use it: Paid developers and organizations; regions and surfaces supported by Google AI Studio/API and Gemini Enterprise Agent Platform.

    How to use it: Open the Gemini 4 Carbon model connection in Google AI Studio or use `gemini-4-carbon` as the model ID in API calls.

    Tip: For long, multi-file tasks, use the long-context workflow with a 1.048.576 combined token limit; remember that the model’s thinking tokens are also billed.

  • ModelGemini Enterprise Agent Platform

    Gemini 4 Carbon for cyber reaches GA for defensive cybersecurity

    Gemini 4 Carbon for cyber has reached GA for defensive tasks such as vulnerability discovery, reverse engineering, exploit validation, and patch generation. The model is offered behind an allowlist through the Fairwind Program, and access to previous cyber models is not transferred automatically.

    Usage details

    Who can use it: Gemini Enterprise Agent Platform; global and us regions through the Fairwind Program or a Google team allowlist.

    How to use it: Apply for access through the Google team or Fairwind Program, then use the `gemini-4-carbon-cyber-permissive` model ID in Agent Platform calls.

    Tip: Run the model on cyber tasks only within an authorized, defensive scope, with human review and logging controls.

  • Pricing / PlanGemini API / Google AI Studio

    Introductory Gemini 4 Carbon API pricing announced through November 30

    Gemini 4 Carbon is available only on the paid tier: through November 30, 2026, input tokens cost $2 per 1 million and output tokens cost $10. Starting December 1, these prices will rise to 4 and 20, respectively; implicit cache pricing will also increase from 0,10 to 0,20.

    Usage details

    Who can use it: Gemini API/Google AI Studio paid tier; Gemini 4 Carbon is not available on the free tier.

    How to use it: Enable a paid project and billing in AI Studio, create an API key, and select the `gemini-4-carbon` model.

    Tip: Estimate production volume using the December 1 prices as well; compare trials through November 30 with the ongoing cost afterward.

  • AvailabilityGemini Enterprise

    Gemini 4 Carbon comes to Gemini Enterprise apps and tools

    Gemini 4 Carbon is now GA in the Gemini Enterprise web and mobile apps, Workflow Builder, and AI developer tools. The model is off by default in the web and mobile apps; organizations using a custom model list need an administrator to select it.

    Usage details

    Who can use it: Gemini Enterprise global, us, and eu multi-regions; web, mobile, Workflow Builder, and licensed AI developer tools.

    How to use it: As an administrator, enable the model in the Manage features on the web app section of the Gemini Enterprise Cloud Console; select Carbon in Workflow Builder and add it to the custom AI developer tools model list.

    Tip: Check the model’s data processing policy before choosing global, us, or eu based on your data residency requirements.

  • AvailabilityGemini app / Google Antigravity / AI Mode in Search

    Gemini 4 Carbon rolling out to Ultra subscribers in the app and AI Mode

    Gemini 4 Carbon has begun rolling out to Google AI Ultra subscribers in the Gemini app, Google Antigravity, and AI Mode in Search. The same announcement also confirms API/AI Studio access for paid developers and organizations.

    Usage details

    Who can use it: Google AI Ultra subscribers; Gemini app, Google Antigravity, and AI Mode in Search. The rollout may be gradual.

    How to use it: Using an Ultra account, open the Gemini app or AI Mode in Search and select Gemini 4 Carbon if it appears in the model picker; use the Carbon connection in AI Studio for developer access.

    Tip: Choose between Argon for everyday fast tasks and Carbon for long verification and multi-step tasks.

  • FeatureGemini Call Screen

    Gemini Call Screen brings conversational screening and live transcription to Pixel 11

    Gemini Call Screen is becoming available on the Pixel 11 series, asking unknown callers why they are calling through a more natural dialogue. Users can follow the conversation in the live transcript on screen and decide when to answer or end the call.

    Usage details

    Who can use it: Pixel 11, Pixel 11 Pro, Pixel 11 Pro XL, and Pixel 11 Pro Fold; in the US.

    How to use it: Tap Call Screen for a call from an unknown number, follow Gemini’s questions and the live transcript, and take over the call when needed.

    Tip: Check critical information provided by the caller in the live transcript before taking personal or financial action.

  • AgentGemini for healthcare

    Gemini for healthcare opens for preview in the U.S.

    Gemini for healthcare brings clinical, financial and administrative systems together through Skills, connectors and a managed Gemini agent, handling multi-step tasks such as medical record synthesis and prior authorization review with human oversight. The solution launched in preview in the U.S. with a governance layer that supports HIPAA compliance.

    Usage details

    Who can use it: Healthcare systems, health plans and payer organizations in the U.S.; preview, invitation and enterprise evaluation requirements apply.

    How to use it: Contact the Google Cloud team through the Gemini for healthcare product page to evaluate the preview; then configure authorized EHR, claims and policy connectors with governance controls.

    Tip: Start the first pilot with a narrow workflow that uses read-only data access, traceable citations and human approval.

  • AvailabilityGemini app

    Gemini task automation expands to the United Kingdom, India and Japan

    Gemini’s multi-step app tasks, such as ordering food, booking a ride and making reservations, are expanding to the United Kingdom, India and Japan. The feature is rolling out with limited access based on supported apps and country requirements.

    Usage details

    Who can use it: Gemini app; selected countries, languages and users aged 18+ in the United Kingdom, India and Japan.

    How to use it: Make a supported food, transportation or restaurant request in natural language in the Gemini app; approve the suggested action and app steps.

    Tip: Always verify the app, person and date Gemini selects for final steps such as payment or reservations.

  • FeatureCreate My Widget

    Create a personal home screen widget with natural language in Gemini

    Create My Widget lets you start designing a personal widget using natural language, without writing code. Widgets can draw on Gmail, Calendar, and Search information to create content such as a travel dashboard, monthly spending tracker, or race countdown.

    Usage details

    Who can use it: Select Pixel devices, countries, and languages; rolling out over the coming weeks. 18+ and some usage limits apply.

    How to use it: Open the Create My Widget flow on your Pixel home screen, describe the widget you want in natural language, and add the generated draft to your home screen.

    Tip: Clearly specify the source and refresh expectations in your widget request; check how personal data is displayed before adding it.

  • AvailabilityGemini Intelligence

    Gemini Intelligence and Rambler expand to Pixel 10 and Motorola Signature 27

    Gemini Intelligence is expanding to the Pixel 10 family and Motorola’s Signature 27 device. Rambler turns spoken thoughts into polished text by removing filler words; U.S. preorders for Motorola Signature 27 begin on November 19.

    Usage details

    Who can use it: Pixel 10, Pixel 10 Pro XL, Pixel 10 Pro Fold and Motorola Signature 27; selected country, language and age requirements apply.

    How to use it: Open the Gemini Intelligence/Gboard flow on a supported device, speak naturally and review Rambler’s edited text before sending it.

    Tip: Check Rambler’s output, especially proper nouns and numbers, before sending.

  • AvailabilityGemini in Gmail

    Gemini summaries and non-personalized writing help in Gmail expand to all Workspace users

    AI Overview summaries of long email threads and non-personalized Help me write in Gemini in Gmail are expanding to more Workspace users aged 18 and older. Features are on by default, can be turned off by users, and administrators can also disable Gemini access.

    Usage details

    Who can use it: All Google Workspace customers, Workspace Individual and personal accounts; users must be 18+ and Gmail smart features must be enabled. The rollout began on November 10.

    How to use it: Open the AI Overview summary above a long email thread in Gmail, or select Help me write in the compose box and enter your prompt; administrators can manage access in the Gemini access settings in Admin console.

    Tip: Use the summary for a quick scan, and verify important dates and obligations in the original emails.

  • APIBigQuery generative AI functions

    BigQuery generative AI functions support Gemini 4 Carbon

    Support for `gemini-4-carbon` has been added to BigQuery’s generative AI functions feature, and the model has been marked GA. This lets generative AI functions in BigQuery run with Carbon.

    Usage details

    Who can use it: BigQuery generative AI functions users; regions and billing conditions supported by Google Cloud.

    How to use it: Update the model selection in the BigQuery generative AI function call you use to `gemini-4-carbon`, then run the query.

    Tip: Measure the same query with the faster Flash model as well to compare cost and latency.

    ✓ Verified by official sourceModels: Gemini 4 CarbonSource: Google Cloud sürüm notları — 10 Kasım 2026
  • OtherCodeMender

    CodeMender v0.20.0 improves large-repository validation and permission errors

    CodeMender v0.20.0 fixes patch application validation in large repositories and with mixed LF/CRLF line endings. It now explicitly reports permission-denied errors in unreadable directories instead of hiding them as zero findings.

    Usage details

    Who can use it: Gemini Enterprise Agent Platform CodeMender users; environments upgrading to v0.20.0.

    How to use it: Upgrade your CodeMender installation to v0.20.0 and revalidate repositories with large files or mixed line endings in `cm verify` runs.

    Tip: Check scan permission errors in the logs before treating a zero-findings result in CI as a success.

  • FeatureGemini CLI

    Gemini CLI `/model` command now supports direct model names and validation

    Gemini CLI v0.69.0-preview.0 lets you pass an argument directly to the `/model` command and validates model names. This reduces menu navigation and typos when switching models.

    Usage details

    Who can use it: Gemini CLI v0.69.0-preview.0 and newer preview releases.

    How to use it: In the CLI, add the model name you want to use directly to the `/model` command; invalid names return a validation error.

    Tip: When pinning a model name in automation scripts, keep the command’s validation output in your CI logs.

  • SettingGemini CLI

    Gemini CLI API header timeout is now configurable

    Gemini CLI v0.69.0-preview.0 adds an option to override the 60-second API header timeout in all authentication modes. You can use the `model.requestTimeoutMs` setting or the `GEMINI_API_HEADERS_TIMEOUT_MS` environment variable.

    Usage details

    Who can use it: Gemini CLI v0.69.0-preview.0 and newer versions; all authentication modes.

    How to use it: Set `model.requestTimeoutMs` in the configuration, or define the `GEMINI_API_HEADERS_TIMEOUT_MS` environment variable before running the CLI.

    Tip: Increase the timeout carefully and only on slow networks; don’t use a high value to mask network problems.

  • FeatureGemini CLI

    Gemini CLI adds `/stats heatmap` for daily token usage

    Gemini CLI v0.69.0-preview.0 adds the `/stats heatmap` subcommand, which displays daily token usage as a heatmap. The feature makes it easier to view usage intensity by day.

    Usage details

    Who can use it: Gemini CLI v0.69.0-preview.0 and newer preview releases.

    How to use it: Run `/stats heatmap` in Gemini CLI.

    Tip: Check the heatmap regularly to spot high-usage days and quota pressure.

  • CapabilityDevice Assistance

    Gemini manages Android settings in natural language with Device Assistance

    Device Assistance lets you ask Gemini to perform tasks such as clearing device storage, checking which apps have location access, or silencing distractions while keeping important calls. The feature is described as a device-aware personal support layer.

    Usage details

    Who can use it: Select Android devices running Android 17 or newer; availability varies by device, language, and country.

    How to use it: Tell Gemini directly what you want it to do on your device; for example, ask it to clear storage or check location access.

    Tip: For tasks that change system settings, check the target app and permission scope before approving the change Gemini suggests.

  • CapabilityGemini on Pixel

    Gemini summarizes long conversations on Pixel in six new languages

    Long conversation summaries on Pixel now support German, French, Spanish, Italian, Hindi and Traditional Chinese. The same capability is also planned for compatible Pixel Watch devices.

    Usage details

    Who can use it: Pixel Watch 3/4/5 and compatible Pixel phone combinations; the six listed languages, Android and device-version requirements apply.

    How to use it: Open a long conversation on a supported device in a supported language; review the summary generated by Gemini on the conversation screen.

    Tip: Use the summary as a quick starting point, not as the source of truth for decisions; verify important details in the original conversation.

  • CapabilityGemini Intelligence / Android Auto

    Gemini’s proactive suggestions coming soon to Pixel 10 and Android Auto

    Gemini Intelligence on Pixel 10 offers timely suggested actions, such as sharing an ETA or finding a dinner spot. Google also announced that Gemini Intelligence on Pixel phones will bring personalized suggestions to Android Auto in the coming weeks.

    Usage details

    Who can use it: Proactive suggestions on Pixel 10 and selected devices; Android Auto integration coming soon. Country, language, age and setup requirements apply.

    How to use it: Use suggestion cards with Gemini Intelligence enabled on Pixel; once the Android Auto rollout arrives, install the Gemini app and complete the required setup.

    Tip: Check calendar, address and contact details before applying a proactive suggestion with one tap.

  • CapabilityGemini on Pixel

    Gemini calling businesses and taking action on Pixel reported

    9to5Google reports that Gemini on Pixel 10 and later can call businesses in the U.S. on a user’s behalf to change reservations, hold products or check inventory. The official Google Pixel Drop and Gemini Intelligence pages I reviewed confirm multi-step task automation but not the phone-calling details; the report is therefore marked unverified.

    Usage details

    Who can use it: According only to 9to5Google’s secondary report, Pixel 10 and later in the U.S.; no official Google confirmation is available.

    How to use it: In the scenario reported by 9to5Google, ask Gemini to call the business and handle the reservation or inventory request; confirm the target business and requested action before the call begins.

    Tip: For transactions conducted by phone, reconfirm the price, date and reservation code Gemini provides with a person.

  • OtherGemini CLI

    Gemini CLI 0.68.0 shows a countdown for 429 quota wait times

    Gemini CLI v0.68.0 displays the cooldown period and countdown for 429 rate-limit responses in the loading indicator. This lets users see how long they need to wait before retrying.

    Usage details

    Who can use it: Gemini CLI v0.68.0 and newer versions.

    How to use it: Upgrade the CLI to v0.68.0 or newer; when a quota response appears, watch the 429 countdown in the loading indicator.

    Tip: While the countdown is running, wait for the current session to retry automatically instead of starting a new session.

    ✓ Verified by official sourceSource: Gemini CLI v0.68.0 sürüm notları
  • SettingGemini CLI

    Gemini CLI can read custom themes from a standalone `theme.json` file

    Gemini CLI v0.69.0-preview.0 supports a standalone `theme.json` configuration for custom themes. Separating theme settings from the main configuration makes them easier to share and move.

    Usage details

    Who can use it: Gemini CLI v0.69.0-preview.0 and newer preview releases.

    How to use it: Put the custom theme definition in a supported `theme.json` file, then select that file in the Gemini CLI theme settings.

    Tip: Keep the `theme.json` file in version control for a consistent look across the team.

  • AvailabilityGemini Enterprise Agent Platform

    Gemini 3.8 Flash rolls out to regional endpoints in Canada and Japan

    Gemini 3.8 Flash is now GA in the `northamerica-northeast1` region in Canada and the `asia-northeast1` region in Japan. Both regions support regional endpoints, in-region machine learning processing, and Provisioned Throughput.

    Usage details

    Who can use it: Gemini Enterprise Agent Platform users in Canada and Japan; the relevant regional endpoints.

    How to use it: Send requests to the `northamerica-northeast1` regional endpoint for Canada or the `asia-northeast1` regional endpoint for Japan, and use `gemini-3.8-flash` as the model.

    Tip: If data residency or latency matters, choose a regional endpoint close to the user instead of the global endpoint.

  • AvailabilityGoogle Workspace / Workspace Studio

    Workspace Skills rolls out to Studio and core Workspace apps

    Workspace Skills—reusable instructions that apply your team’s rules, templates, and Drive references to Gemini—has begun rolling out to Workspace Studio and the Gmail, Docs, Slides, Drive, and Chat surfaces. Users can create, test, and share Skills on the web, and run active Skills automatically or with `@`.

    Usage details

    Who can use it: Business Starter/Standard/Plus, Enterprise Starter/Standard/Plus, Google AI Pro for Education, and AI Expanded Access; ages 18+; English only; creating on the web and invoking in Chat on mobile are supported, while Sheets/Meet are not yet available. Rapid rollout begins November 9; Scheduled rollout begins November 30.

    How to use it: Create, test, and share Skills in Workspace Studio > Skills; in Gmail, Docs, Slides, Drive, or Chat, tell Gemini what you need or invoke one or more Skills with `@`.

    Tip: Keep enterprise rules in SKILL.md concise and testable; note that Gemini app Skills and Workspace Skills do not sync automatically.

  • FeatureGoogle Chat

    Gemini schedules meetings from Chat conversations with one click

    In Chat conversations with up to five people, Gemini can detect a clear meeting request, underline the text and show suggested times, participants and a title. Users can select a time and click Schedule to create a meeting without leaving the conversation.

    Usage details

    Who can use it: Google Chat on the web and Chat in Gmail; initially in English, for Business Standard/Plus, Enterprise Standard/Plus, Frontline Plus and AI Ultra Access.

    How to use it: Hover over the underlined text in a message suggesting a meeting, check the suggested time and title, then click Schedule; select Open in Calendar if you need more details.

    Tip: State a clear meeting purpose and duration in the message to help Gemini detect it correctly.

  • FeatureWorkspace Studio

    Workspace Studio flows get a `Wait until` step for up to 14 days

    The new `Wait until` step can resume a flow at a specified date and time or after a selected duration; it can also be added inside condition branches and Repeat for each loops. This makes it possible to set up delayed emails, reminders, and Chat summaries in a single flow.

    Usage details

    Who can use it: Business Starter/Standard/Plus, Enterprise Standard/Plus, Education editions, and AI add-ons; full rollout began November 9.

    How to use it: Add a `Wait until` step in the Workspace Studio flow editor, choose a specific date and time or duration, and track the wait and resume record in the Activity tab.

    Tip: Because the maximum is 14 days, plan the workflow duration before splitting longer waits into a separate flow.

OpenAI · ChatGPT

#4 GPT-6 Astra#5 GPT-6.1 Sol#7 GPT-6.1 Terra
Full history (339) →

New in this edition · 34

  • RetiredChatGPT

    Advanced and Standard Voice will be retired on December 11

    OpenAI announced that it will remove the Advanced and Standard voice options from ChatGPT Voice on December 11, 2026; they remain available during the transition period. After that date, selections will be routed to GPT-Live, but live video and screen sharing will continue to use Advanced voice for now.

    Usage details

    Who can use it: Accounts using ChatGPT Voice; notifications began on November 10, and removal will apply to all affected users on December 11, 2026.

    How to use it: During the transition period, you can make a selection under Settings > Voice; after December 11, GPT-Live will be the default voice.

    Tip: Test workflows that rely on custom voice behavior with GPT-Live now and adjust expectations before the transition.

  • RetiredOpenAI API

    gpt-image-2 and its snapshot deprecated in the API

    `gpt-image-2` and the `gpt-image-2-2026-04-21` snapshot have been deprecated and will be removed from the API on February 10, 2027. OpenAI recommends the `gpt-image-2.6-sunburst` or `gpt-image-2.6-flare` models for migration.

    Usage details

    Who can use it: All developers using the OpenAI API; removal date: 2027-02-10.

    How to use it: Replace the old model IDs in your code and configurations with 2.6 Sunburst or 2.6 Flare, then repeat your image generation and editing tests.

    Tip: Update all workflows that pin the model ID and rerun cost tests before the removal date.

    ✓ Verified by official sourceModels: gpt-image-2, gpt-image-2-2026-04-21, gpt-image-2.6-sunburst, gpt-image-2.6-flareSource: OpenAI API: Deprecations
  • FeatureChatGPT

    Businesses in ChatGPT opens in beta

    The Businesses in ChatGPT beta lets businesses claim their profiles through Business Hub, update their information and product feeds, and view aggregate insights about interactions in ChatGPT.

    Usage details

    Who can use it: Eligible businesses in locations where ChatGPT is available; Enterprise customers and advertisers can apply through their OpenAI representatives, while Shopify merchants can apply through the Shopify connection.

    How to use it: Open Business Hub and verify your business; update basic information and product price and inventory feeds, then review the insights section once sufficient interactions have accumulated.

    Tip: Keep your business information and product feed up to date to make your visibility and click insights in ChatGPT more meaningful.

  • AgentChatGPT

    Business Agents launches in the US for Businesses in ChatGPT

    Eligible e-commerce and online service businesses can create, test, and publish a Business Agent after verifying their business. Users can start a separate, labeled business conversation from the business profile in a ChatGPT response.

    Usage details

    Who can use it: Beta; limited to eligible e-commerce and online service businesses in the United States.

    How to use it: Verify your business in Business Hub, then create, test, and publish the Business Agent.

    Tip: Keep catalog, pricing, and availability information synchronized with your business source to ensure reliable Agent responses.

  • SettingCodex CLI

    Choose a model and reasoning effort in Codex Cloud conversations

    Cloud conversations now retrieve the account-scoped `/v2/models` catalog and let you select a model and reasoning effort for a new turn. The selection is sent only with the next new turn, and the user is notified when it is accepted.

    Usage details

    Who can use it: Codex CLI 0.169.0-alpha.10 and later; Cloud servers that support model discovery.

    How to use it: Open the model picker in a Cloud conversation, select a supported model and reasoning effort, then send the next turn.

    Tip: For long tasks, start with a lower effort for exploration and use a higher effort only for critical turns.

  • APICodex app-server

    Codex app-server changes environments through turn and thread updates

    `turn/settings/update` and `thread/settings/update` now accept the experimental `environments` fields. A turn update applies to subsequent steps, while a thread update applies to future turns; an empty array clears the selection, and `null` preserves the current selection.

    Usage details

    Who can use it: Codex app-server 0.169.0-alpha.10 and later; users of the experimental settings API.

    How to use it: Send an ordered `environments` list in the relevant settings update request; use `[]` to clear it and `null` to preserve it.

    Tip: Select an environment at the thread level for future turns without affecting the running turn.

  • APICodex app-server

    Codex can sample in-progress turns

    `thread/sample` can now use `allowInProgressTurn` to take a snapshot of the current state before the source turn completes. The source continues running; pending tool calls may appear as `aborted` in the snapshot.

    Usage details

    Who can use it: Codex app-server 0.169.0-alpha.9 and later.

    How to use it: Add `allowInProgressTurn: true` to the `thread/sample` request; make sure the sampled turn is the latest turn in the source.

    Tip: Use it to view live progress in long workflows, but don't treat the sample as the final result.

  • SettingCodex CLI

    Codex CLI can summarize and collapse the full-screen transcript

    The full-screen transcript view now includes output-free summaries for tool calls, consecutive call groups, and the ability to collapse completed-turn activity. `tui.collapse_turn_details` is off by default and can be enabled using the Collapse turn details option in full-screen settings.

    Usage details

    Who can use it: Codex CLI 0.169.0-alpha.10 and later.

    How to use it: Open the full-screen transcript settings and enable Collapse turn details; expand summary rows to return to the details.

    Tip: In long sessions, use the collapsed view; when investigating errors, use search and the expanded details view.

  • FeatureCodex CLI

    Codex CLI can save a diagnostic ZIP locally

    In the TUI report note editor, `Ctrl+S` saves the diagnostic ZIP to local disk without sending it to OpenAI; the saved archive's path is shown when the operation completes. The feature requires a local app-server and can also work with uploads disabled.

    Usage details

    Who can use it: Codex CLI 0.169.0-alpha.9 and later; sessions with a local app-server.

    How to use it: Open the report note editor in the TUI and use `Ctrl+S`; the `Enter` behavior for submission remains unchanged.

    Tip: Review the local ZIP's contents and sensitive logs before sending it to the support team.

  • FeatureCodex CLI

    Archive and restore Codex Cloud threads

    Cloud threads can now be archived, archived records can be listed, and threads can be restored. Archiving preserves the local session that shares the same ID and does not affect unselected child agents.

    Usage details

    Who can use it: Codex CLI 0.169.0-alpha.10 and later; Cloud conversations.

    How to use it: Use the archive action for the relevant thread in the Cloud conversation resume picker; restore it from the list of archived records.

    Tip: Archive completed work, but check the status of a thread with active child agents before archiving it.

  • APICodex app-server

    MCP content helper added for Codex Code Mode

    Code Mode can pass text, image, and audio blocks from MCP tool results using `content(block)` and `content(iterable)`, preserving their type and order. Source links and embedded resources are serialized as text; invalid or encrypted blocks are rejected.

    Usage details

    Who can use it: Codex CLI 0.169.0-alpha.10 and later; Code Mode MCP integrations.

    How to use it: Pass the tool result to Code Mode using `content(result.content)` directly, or `content(iterable)` for iterable blocks.

    Tip: For mixed MCP outputs, test the distinction between text, images, and audio on the client side while preserving block order.

  • AvailabilityCodex CLI

    Device-code sign-in for the Codex headless TUI

    When supported by the gateway, Codex TUI prefers device authentication in SSH, headless Linux, or sessions without a local browser. The interface displays a verification URL, a one-time code, and cancellation instructions.

    Usage details

    Who can use it: Codex CLI 0.169.0-alpha.10 and later; gateway providers that support device authorization.

    How to use it: Select the relevant gateway provider in a headless session; complete browser verification using the URL and one-time code shown in the TUI.

    Tip: Device login may not be supported on older gateway servers, so upgrade the app-server to a current version as well.

  • RetiredCodex CLI

    Two options removed from Codex managed Guardian settings

    When using managed policy, `auto_review.extra_policy` and `auto_review.experimental_policy_template` have been removed from the configuration and schema. The recommended path for additional rules is `guardian_extra_policy` in the managed `requirements.toml`.

    Usage details

    Who can use it: Codex CLI 0.169.0-alpha.10 and later; installations using managed Guardian policy.

    How to use it: Remove the two old fields from the configuration and define additional rules under `guardian_extra_policy` in the managed `requirements.toml`.

    Tip: Before upgrading to the alpha, scan for the old fields and avoid treating warnings in CI logs as errors.

  • SettingCodex CLI

    Codex distinguishes read-only MCP tools in auto-approval

    When an MCP tool has `readOnlyHint: true`, this flag now takes precedence in auto-approval decisions, even if it conflicts with `destructiveHint` or `openWorldHint`. Existing approval checks remain in place for open-prompt and strict automatic review modes.

    Usage details

    Who can use it: Codex CLI 0.169.0-alpha.10 and later; MCP auto-approval flows.

    How to use it: Add `readOnlyHint: true` to the descriptions of read-only MCP tools; keep the appropriate approval mode for write tools.

    Tip: Separating read-only and modifying tools with accurate annotations reduces unnecessary approvals.

  • CapabilityCodex CLI

    Codex extends regular Responses tool notifications to Bedrock and GCP

    With `incremental_tools_legacy`, tool definitions can still be sent iteratively in regular Responses streams; support was added for Bedrock providers in 0.167.1 and for GCP Responses requests in the alpha stream. On GCP, definitions are sent via `additional_tools`.

    Usage details

    Who can use it: Codex CLI 0.167.1 and 0.169.0-alpha.9 and later; Amazon Bedrock and GCP providers.

    How to use it: Enable `features.incremental_tools_legacy = true` in Codex features and use a supported Bedrock or GCP provider.

    Tip: Enable it to reduce network load for large tool catalogs; first verify the provider's incremental tool behavior in a test session.

  • Plugin / IntegrationCodex app-server

    OAuth metadata discovery comes to the Codex RMCP client

    The RMCP client exposes a `discover_authorization_metadata` call using the provided HTTP client and can discover authorization server information through protected-resource metadata. It returns `None` when no metadata is published and reports discovery errors.

    Usage details

    Who can use it: Codex app-server 0.169.0-alpha.9 and later; OAuth-enabled RMCP connections.

    How to use it: In the RMCP client, use the `discover_authorization_metadata` call with the existing OAuth HTTP adapter and calling client.

    Tip: Verify published issuer and endpoint metadata before sending credentials.

  • Plugin / IntegrationCodex CLI

    OpenSSH ProxyCommand mode for Codex tcp-tunnel

    `codex tcp-tunnel` now has a mode for OpenSSH `ProxyCommand` that carries SSH sessions over HTTP/3 CONNECT. Use `--environment-id` and `--port` to select the environment and target port; optional `--proxy-url` and `--target` are also supported.

    Usage details

    Who can use it: Codex CLI 0.169.0-alpha.10 and later; requires ChatGPT OAuth, cloud configuration, and a suitable network policy.

    How to use it: Run `codex tcp-tunnel` with `--environment-id ENV --port PORT` and define it as ProxyCommand in your OpenSSH configuration.

    Tip: Leave SSH authentication and host-key verification to OpenSSH rather than Codex.

  • Plugin / IntegrationCodex CLI

    Codex TUI app-server uses MCP configuration

    The TUI reads the active configuration of MCP servers through `config/read` and syncs startup events with progress. This reduces startup indicators that remain pending and namespace conflicts when local and app-server MCP lists differ.

    Usage details

    Who can use it: Codex CLI 0.169.0-alpha.9 and later; sessions using app-server MCP.

    How to use it: Update the CLI; use the app-server's `config/read` response as the source of truth instead of copying MCP configuration separately to the client.

    Tip: After local MCP changes, start a new session so the startup state is read again.

  • APIOpenAI Node SDK

    Node SDK adds support for typed async function tools

    OpenAI Node SDK v7.55.0 adds support for typed async function tools, enabling typed use of asynchronous function tool definitions.

    Usage details

    Who can use it: Node SDK v7.55.0 and later; JavaScript or TypeScript applications implementing function tools.

    How to use it: Upgrade the package to v7.55.0 and use your typed async function tool definition in an Agents or Responses call.

    Tip: Keep TypeScript types aligned with the runtime schema to reduce tool-calling errors in production.

  • APIOpenAI Python SDK

    Python SDK adds support for typed async function tools

    OpenAI Python SDK v3.49.0 adds typed async support to function tool definitions; the SDK can carry the schema and call types for asynchronous tools.

    Usage details

    Who can use it: Python SDK v3.49.0 and later; API applications that use function tools.

    How to use it: Upgrade the `openai` package to v3.49.0 and connect your typed async function tool definition to your existing Agents or Responses flow.

    Tip: Explicitly type tool arguments and return values, and test for unexpected asynchronous results.

  • APIOpenAI Ruby SDK

    Ruby SDK types Managed Agents async fields

    OpenAI Ruby SDK v0.122.0 adds typed async fields for Managed Agents function tools. This lets Ruby represent the asynchronous configuration of these tools with explicit types.

    Usage details

    Who can use it: Ruby SDK v0.122.0 and later; applications that use Managed Agents function tools.

    How to use it: Upgrade the gem to v0.122.0 and use the typed async fields in Managed Agents function tool definitions.

    Tip: Pin the gem version in your application lockfile and add end-to-end tests for asynchronous tools.

  • FeatureCodex CLI

    Floating shortcut completion guide for Codex CLI

    Instead of footer hints, Codex displays possible completions for pending key combinations in a floating guide after about 350 ms. The guide works across chat, transcript, session selector, draft, and confirmation views.

    Usage details

    Who can use it: Codex CLI 0.169.0-alpha.10 and later.

    How to use it: After pressing a shortcut prefix, wait about 350 ms; press `Esc` to cancel the completion.

    Tip: In narrow terminal windows, wait for the guide to appear to see which action will be completed.

  • OtherCodex CLI

    Codex CLI fixes Windows quoted-path and PowerShell directory issues

    In Windows sandbox `cmd /c` calls, quoted paths and executable names are now preserved correctly; the MXC PowerShell sandbox also restores the `$PWD` location that was lost at startup. This makes paths containing spaces and relative file access more reliable.

    Usage details

    Who can use it: Codex CLI 0.169.0-alpha.10 and later; sessions using the Windows sandbox and PowerShell.

    How to use it: Update the CLI; rerun Windows paths containing spaces and PowerShell commands.

    Tip: When debugging regressions, first check that the working directory and quoted path appear correctly in the CLI logs.

  • Plugin / IntegrationCodex CLI

    Codex CLI recognizes local Claude Code skill migration

    Codex recognizes `claude-code-local-skills` as a separate migration type; in local mode, it does not import the `synced` folder and excludes top-level or nested `.trash` folders.

    Usage details

    Who can use it: Codex CLI 0.169.0-alpha.9 and later; users migrating Claude Code skills.

    How to use it: Select the `claude-code-local-skills` source in the skill import or migration flow; keep any skills you want to retain outside `.trash`.

    Tip: Use the default migration mode instead of local-only if you want to retain synced skills.

  • FeatureCodex CLI

    Codex adds a reporting option to the safety refusal flow

    After a safety check refusal, Codex shows the status while preparing alternatives and offers an option to report the refusal. It also makes clearer that choosing rewrite starts a new chat with the previous context.

    Usage details

    Who can use it: Codex CLI 0.169.0-alpha.9 and later; flows with the reporting feature enabled.

    How to use it: Use the available report option on the safety refusal screen; if you choose rewrite for an alternative, confirm the new chat flow.

    Tip: Include context in false-positive reports to make it easier for the safety team to review them.

  • FeatureCodex app-server

    Codex preserves first-party order in the model catalog

    First-party model catalogs from the OpenAI provider, whether delivered through the Codex backend or `model_catalog_url`, now preserve catalog order in the display. Priority-based default model selection and third-party catalog behavior remain unchanged.

    Usage details

    Who can use it: Codex app-server 0.169.0-alpha.9 and later; OpenAI provider catalogs.

    How to use it: Use an updated app-server that presents the model catalog in the order returned by the provider; manage the priority setting separately for the default model.

    Tip: To change the order of models shown to users, adjust catalog order; to change the default, handle the priority value separately.

  • OtherOpenAI Python SDK

    Python SDK encodes microphone recordings as valid PCM WAV

    Python SDK v3.49.0 fixes the encoding of microphone recordings as valid PCM WAV. This reduces format-related rejections or playback issues when sending audio input to the API.

    Usage details

    Who can use it: OpenAI Python SDK v3.49.0 and later; applications that send microphone recordings.

    How to use it: Upgrade the SDK to v3.49.0 and rerun your existing microphone recording flow without changes.

    Tip: If needed, also record the sample rate and channel count to make it easier to compare problematic devices.

  • CapabilityCodex CLI

    Codex CLI preserves image orientation when pasting on macOS

    On macOS, Codex now prioritizes suitable native PNG or JPEG clipboard data, reducing unnecessary TIFF conversion and applying EXIF orientation information to pasted images. This behavior also extends to transcript thumbnails.

    Usage details

    Who can use it: macOS Codex CLI 0.169.0-alpha.7 and later.

    How to use it: Update the CLI and paste an image directly from the clipboard into a Codex conversation on macOS.

    Tip: For phone photos whose orientation is determined by metadata, try pasting directly from the clipboard instead of using the file first.

  • AvailabilityCodex app-server

    Codex gateway adds device authorization support for OAuth

    Codex app-server gateway OAuth can now report device authorization capability and start sign-in with `deviceAuth: true`. The flow returns a verification URL, a one-time code, and a paired `loginId`, then sends the result to the connection that initiated it.

    Usage details

    Who can use it: Codex gateway OAuth servers that support device authorization; 0.169.0-alpha.7 and later.

    How to use it: Check the `supportsDeviceAuth` value in the `account/gatewayOAuth/read` response, then send `deviceAuth: true` in the `account/gatewayOAuth/login` request and complete the flow using the URL and code.

    Tip: Use the returned `prompt` field to avoid requesting device authorization again if the saved credentials are still valid.

  • APICodex app-server

    Codex Realtime attachments add cancellation and deadline options

    Codex adds `RealtimeAttachmentOptions` to RealtimeConversationAttach operations, with a cancellation receiver and a `start_before` deadline. Pending connection handshakes can now be canceled or cleanly terminated when they time out.

    Usage details

    Who can use it: Developers using the Codex app-server protocol; 0.169.0-alpha.7 and later.

    How to use it: Pass `RealtimeAttachmentOptions` to the `Op::RealtimeConversationAttach` call and set the cancellation receiver and `start_before` time limit.

    Tip: When a user selects a new attachment, cancel the previous pending connection to prevent stale requests from piling up in the queue.

  • SettingCodex CLI

    Codex passes untrusted context as function_call_output

    `untrusted_as_tool_output` is now stable and enabled by default; additional context such as browser information and targets is now passed to the model as `function_call_output` items in the `agent` namespace instead of as a user message. The setting can be disabled to preserve the old format.

    Usage details

    Who can use it: Codex CLI 0.169.0-alpha.7 and later.

    How to use it: To revert to the old message format, add the `features.untrusted_as_tool_output = false` setting to your configuration.

    Tip: Update client parsers that process tool context as separate function outputs to work with the new default.

  • APIOpenAI Go SDK

    Go SDK recognizes gpt-6-rosalind in beta Responses

    OpenAI Go SDK v3.98.0 made the `gpt-6-rosalind` model constant available in beta Responses requests. This completes beta Responses type support, separate from its addition to the model catalog.

    Usage details

    Who can use it: Go SDK v3.98.0 and later; organizations with access to the beta Responses API.

    How to use it: Upgrade the SDK to v3.98.0 and select `gpt-6-rosalind` as the model in a beta Responses request.

    Tip: Use the SDK constant for the model name instead of raw text to reduce typos and version incompatibility.

    ✓ Verified by official sourceModels: gpt-6-rosalindSource: OpenAI Go SDK v3.98.0 sürüm notları
  • APIOpenAI Python SDK

    Python SDK adds support for gpt-6-rosalind in beta Responses

    OpenAI Python SDK v3.48.0 recognizes the `gpt-6-rosalind` model in beta Responses requests. This adds a beta Responses surface to the existing general model catalog support.

    Usage details

    Who can use it: Python SDK v3.48.0 and later; organizations with access to the beta Responses API.

    How to use it: Upgrade the `openai` package to v3.48.0 and select the `gpt-6-rosalind` model in a beta Responses call.

    Tip: For beta use, verify model access alongside organization and project permissions.

    ✓ Verified by official sourceModels: gpt-6-rosalindSource: OpenAI Python SDK v3.48.0 sürüm notları

Meta · Meta AI

#9 Muse Wonder FullMuse WonderMuse Ember
Full history (128) →

New in this edition · 5

  • AvailabilityMeta AI on AI glasses

    Meta AI on AI glasses is coming to 12 new countries and regions

    Meta AI on screen-free AI glasses is becoming available in Indonesia, Thailand, Portugal, Poland, Israel, Malaysia, New Zealand, Philippines, South Africa, Hong Kong, Turkey and Macau. Access will roll out starting the week of November 11, 2026.

    Usage details

    Who can use it: New regions: Indonesia, Thailand, Portugal, Poland, Israel, Malaysia, New Zealand, Philippines, South Africa, Hong Kong, Turkey and Macau; a compatible AI glasses model and the Meta AI mobile app are required.

    How to use it: Update your glasses and the Meta AI mobile app, pair them with your Meta account, and use Meta AI through your glasses when the rollout begins in your region.

    Tip: If regional access isn’t available, first check the app and firmware versions; try again during the rollout week.

  • AvailabilityMeta AI on AI glasses

    Meta AI on AI glasses expands to nine new languages, including Turkish

    Meta AI support on screen-free AI glasses such as Ray-Ban Meta, Oakley Meta and Meta Glasses is expanding to Arabic, Hebrew, Indonesian, Malaysian, Mandarin, Polish, Portuguese, Tagalog, Thai and Turkish. The changes will roll out starting the week of November 11, 2026.

    Usage details

    Who can use it: Screen-free AI glasses; the new language list includes Arabic, Hebrew, Indonesian, Malaysian, Mandarin, Polish, Portuguese, Tagalog, Thai and Turkish.

    How to use it: Update your AI glasses and the Meta AI mobile app to the latest version; once the update is available, use Meta AI by speaking in a supported language.

    Tip: When using Turkish, give short, direct commands; check the response language and regional feature availability during your first attempts.

  • FeatureInstagram Effect Builder

    Instagram Effect Builder creates Stories effects from prompts

    Effect Builder lets users create AI effects for Instagram Stories using text prompts, without requiring design or coding skills. The workflow includes adding a style reference image, previewing with a photo, follow-up prompts and publishing the effect after a safety review.

    Usage details

    Who can use it: For people aged 18 and older living in the US, through Edits’ web-based tool; created effects can be made available to others on Instagram Stories after approval.

    How to use it: Go to https://edits.com/effectbuilder, describe your effect idea, optionally add a style reference image, preview it with your own photo, then edit it and submit it for a safety review.

    Tip: Start with one clear visual goal; preview with different photos before publishing and use follow-up prompts to correct unwanted results.

  • AvailabilityMuse

    Muse is now available on Apple Watch and Android Wear OS

    Meta’s official Muse account announced that Muse is now available on Apple Watch and Android Wear OS. According to TechCrunch, users can interact with it from their watch through voice search, text messages or dictation.

    Usage details

    Who can use it: Apple Watch and Android Wear OS; the official announcement doesn’t specify countries, plans or a minimum operating system requirement.

    How to use it: Open or install Muse on an Apple Watch or Android Wear OS device; then send your request by voice search, text or dictation.

    Tip: Use short, clear commands on your watch; it’s more practical to complete long texts and multi-step tasks on your phone.

  • APIMeta VR Android Studio Plugin ve Meta VR CLI

    Android Studio Rabbit, headset casting and Meta VR CLI v1.10.0 added

    The November 9 releases of the Meta VR Android Studio Plugin add support for Android Studio Rabbit (2026.2). Meta VR Casting is available from Android Studio’s Physical Device section, and Meta VR CLI is updated to v1.10.0.

    Usage details

    Who can use it: Developers using Meta VR Android Studio Plugin versions 253.2.0.1.77, 261.2.1.0.77 and 262.2.1.0.77; Android Studio Rabbit (2026.2).

    How to use it: Install a compatible version of Meta VR Android Studio Plugin; select Meta VR Casting from the Physical Device section in Android Studio, and use the updated Meta VR CLI v1.10.0 in terminal/AI tool workflows.

    Tip: After switching to Android Studio Rabbit, test the physical device connection and casting workflow separately; also verify CLI-based CI workflows with the updated version.

Moonshot AI · Kimi

#10 Kimi K3.1 Pro
Full history (320) →

New in this edition · 22

  • FeatureKimi Code CLI

    NotifyUser updates appear as colored bullets in fullscreen mode

    In the fullscreen TUI, intermediate `NotifyUser` updates are now shown with a bullet in the primary color instead of the `Update` badge. The change preserves the same heading layout in collapsed and expanded views at narrow widths.

    Usage details

    Who can use it: On the official kimi-code repository’s main branch; not listed separately in the release notes for 2.5.3, released on November 10. Check whether it is available in your installation with `kimi --version` and in subsequent release notes. Fullscreen Kimi Code CLI/TUI sessions using `NotifyUser` intermediate progress updates.

    How to use it: Enable intermediate updates with `KIMI_CODE_EXPERIMENTAL_NOTIFY_USER=1` and use the fullscreen TUI; updates will appear as colored bullets.

    Tip: For long intermediate notifications, the compact bullet in fullscreen view takes up less horizontal space than the `Update` badge.

  • APIKimi Code server API

    `tool_call` items are linked to background tasks with `task_id`

    When a tool call's work starts as a background task or is moved to the background later, `task_id` is added to the `tool_call` item in the conversation stream. The client can use this ID to continue monitoring and stopping the task; foreground calls can be found through the parent tool call.

    Usage details

    Who can use it: On the main branch of the official kimi-code repository; not separately listed in the 2.5.3 release notes published on November 10. Check whether it is available in your installation with `kimi --version` and subsequent release notes. Custom clients using the Kimi Code WebSocket conversation stream and task API.

    How to use it: Read `tool_call.task_id` from the WebSocket stream and send `task.watch`/`task.stop` requests for the task using that ID; map foreground tasks via `parent_tool_call_id`.

    Tip: Do not tie a tool call only to the turn lifecycle; background tasks can continue after a turn ends, so use `task_id` as the persistent mapping key.

  • APIKimi Code server API

    API error envelopes no longer leak stack traces

    The `stack` field has been removed from KAP server error envelopes; missing providers, credentials, or models are now more accurately classified as setup/readiness errors. The `40935` code for cases where Tower mode cannot be entered and the appropriate route error mappings have also been added to the reference.

    Usage details

    Who can use it: On the official kimi-code repository’s main branch; not listed separately in the release notes for 2.5.3, released on November 10. Check whether it is available in your installation with `kimi --version` and in subsequent release notes. Kimi Code server API clients and integrations that handle error envelopes or provider/model setup status.

    How to use it: Rely only on the `code`, `msg`, and `details` fields in error responses; handle setup codes such as `40110`/`40113` when a provider/model is not ready, and no longer expect `stack`.

    Tip: In production clients, use `msg` to display the error to users; do not make internal server details part of the client contract.

  • APIKimi Code server API

    `event.session.restored` is published when an archived session is restored

    When an archived session is restored with `:restore`, the server now publishes a global `event.session.restored` event to all open WebSocket connections. The event carries `workspace_id` and `sessionId` and is sent when the session is actually removed from the archive.

    Usage details

    Who can use it: On the official kimi-code repository’s main branch; not listed separately in the release notes for 2.5.3, released on November 10. Check whether it is available in your installation with `kimi --version` and in subsequent release notes. Kimi Code WebSocket clients and integrations that update session lists through events instead of polling.

    How to use it: Listen for global events on the `/api/v1/ws` connection and, after `POST /api/v1/sessions/{session_id}:restore`, refresh the session cache for the `sessionId` in the `event.session.restored` payload.

    Tip: Don’t expect a session subscription for this global event; design the client to process events from all connections.

  • SettingKimi Code CLI

    With the experimental flag enabled, TUI opens in fullscreen by default

    When `KIMI_CODE_EXPERIMENTAL_FLAG=1` is set and no separate TUI mode is selected, the CLI now prefers fullscreen mode instead of regular mode. An explicit setting of `KIMI_CODE_TUI_FULL_SCREEN` overrides this preference; it falls back to regular mode in incompatible environments such as tmux control mode, Termux, and dumb terminals.

    Usage details

    Who can use it: On the main branch of the official kimi-code repository; not separately listed in the 2.5.3 release notes published on November 10. Check whether it is available in your installation with `kimi --version` and subsequent release notes. Interactive Kimi Code CLI/TUI users in environments where the experimental flag is enabled.

    How to use it: Run `KIMI_CODE_EXPERIMENTAL_FLAG=1 kimi` in the terminal; to force normal mode, use `KIMI_CODE_TUI_FULL_SCREEN=0 kimi`.

    Tip: In CI or narrow terminal environments, explicitly select regular mode with `KIMI_CODE_TUI_FULL_SCREEN=0` instead of relying on automatic fallback.

  • APIKimi Code server API

    Manual compact now returns an explicit error during ongoing compaction

    If a second manual compaction is requested through the Web interface or the `:compact` endpoint while compaction is already running, the request is no longer silently ignored; it returns code `40910` and a retry message. This lets the client determine that the operation did not actually start.

    Usage details

    Who can use it: On the official kimi-code repository’s main branch; not listed separately in the release notes for 2.5.3, released on November 10. Check whether it is available in your installation with `kimi --version` and in subsequent release notes. Kimi Code Web and the local server API; concurrent manual/automatic compact requests in the same session.

    How to use it: Resend `POST /api/v1/sessions/{session_id}:compact` after the current compact finishes; if you receive `40910`, don’t start a new request.

    Tip: On a `40910` response, the client should monitor the ongoing compaction and schedule a single retry, rather than using exponential retry.

  • APIKimi Code server API

    `turn_number` is now the primary parameter for file history queries

    File history change and content endpoints now use the `turn_number` parameter; the old `turn_id` name remains accepted as a deprecated alias for backward compatibility. If no turn number is provided, a validation error is returned.

    Usage details

    Who can use it: On the official kimi-code repository’s main branch; not listed separately in the release notes for 2.5.3, released on November 10. Check whether it is available in your installation with `kimi --version` and in subsequent release notes. Kimi Code server API file history clients: `/file-history/changes` and `/file-history/content`.

    How to use it: Use `GET /api/v1/sessions/{session_id}/file-history/changes?turn_number=1` or `.../content?turn_number=1&path=dosya.txt`.

    Tip: Use `turn_number` instead of `turn_id` in new client code; don’t assume alias support will remain indefinitely for older clients.

  • FeatureKimi Code CLI

    File suggestions now include hidden and gitignored paths

    `@` file completion now suggests paths excluded by `.gitignore` by default; hidden paths are retained, `.git` is excluded, and regular files appear first when match quality is equal. Suggestion search has been streamlined, and existing results can be previewed quickly as characters are added to the typed query.

    Usage details

    Who can use it: On the main branch of the official kimi-code repository; not separately listed in the 2.5.3 release notes published on November 10. Check whether it is available in your installation with `kimi --version` and subsequent release notes. Kimi Code CLI/TUI file-mention completion flow and associated server `fs:suggest` usage; the target filesystem in remote workspaces.

    How to use it: Type `@` in a Kimi Code prompt and search for a hidden file path or one inside `.gitignore`; add the selected suggestion to the prompt.

    Tip: Typing a distinctive beginning of the full path brings the correct file inside gitignore to the top faster thanks to exact/prefix ordering.

  • CapabilityKimi Code

    A pending question in a stopped turn is now cancelled as `turn_ended`

    If the user stops a turn while `AskUserQuestion` is waiting, the pending interaction now closes with `cancelled: true` and `reason: turn_ended` instead of appearing as an unanswered question. This lets the client distinguish between a user response and a stop.

    Usage details

    Who can use it: On the main branch of the official kimi-code repository; not separately listed in the 2.5.3 release notes published on November 10. Check whether it is available in your installation with `kimi --version` and subsequent release notes. Kimi Code Web/TUI and host clients that use the AskUserQuestion interaction.

    How to use it: Use the Stop action while a question is pending; check the `cancelled` and `reason: turn_ended` fields in the interaction resolution.

    Tip: Handle `turn_ended` separately from an explicit user `dismissed` response; do not start an automatic re-questioning loop.

  • AvailabilityKimi Code CLI

    A mandatory upgrade prompt has been added for older CLI versions

    If a remotely fetched client policy reports a `cliMinVersion` higher than the current version, the interactive CLI offers install, retry, or temporary skip options before continuing. If the update target is not yet available in the channel or installation fails, an error state is shown; non-interactive runs skip this prompt.

    Usage details

    Who can use it: On the main branch of the official kimi-code repository; not separately listed in the 2.5.3 release notes published on November 10. Check whether it is available in your installation with `kimi --version` and subsequent release notes. Interactive Kimi Code CLI users running a supported installation source, on versions below the minimum set by policy.

    How to use it: Start `kimi` interactively with an older CLI; on the `Kimi Code Upgrade Required` screen, use the arrow keys to select `Install update now`, `Retry`, or, when permitted, `Skip for this run`, then press Enter.

    Tip: If automatic upgrading is not possible, use the target version and manual installation information on the same screen; continuing with the older version may be blocked again on the next startup.

  • APIKimi Code server API

    Raw file download errors now return actual HTTP 4xx/5xx codes

    File download errors on `GET /api/v1/files/{file_id}`, `/api/v1/fs:content`, and session `:download` endpoints now return actual HTTP 4xx/5xx status codes and the standard error envelope. Clients no longer have to read parameter validation and storage/session errors solely from application codes inside HTTP 200 responses.

    Usage details

    Who can use it: On the main branch of the official kimi-code repository; not listed separately in the release notes for 2.5.3, released on November 10. Check whether your installation includes it with `kimi --version` and subsequent release notes. For Kimi Code local server API clients and integrations using these raw file endpoints.

    How to use it: For each request, check the HTTP status code first, then the `code`, `msg`, and `details` fields in the JSON envelope, if present; handle the 400, 403, 404, 409, 413, and 500 responses documented in OpenAPI.

    Tip: For backward-compatible clients, reading both the actual HTTP status and the envelope code preserves compatibility with older and newer server versions.

  • FeatureKimi Code

    Kimi Code 2.5.3: a steer message interrupts the active model request and continues in the same turn

    In Kimi Code 2.5.3, a steer message sent to an active session now interrupts the current model request, and the work continues with the new direction in the same turn. The `POST /api/v1/sessions/{session_id}/steer` endpoint was also added to send a steer message with a single request.

    Usage details

    Who can use it: Kimi Code 2.5.3 (published on npm on November 10) and later; CLI/TUI, Web, and local server API.

    How to use it: Update with `npm i -g @moonshot-ai/kimi-code@latest` and verify 2.5.3 with `kimi --version`. While the agent is running, write and send your new instruction; to steer from your own client, use the `POST /api/v1/sessions/{session_id}/steer` request.

    Tip: When the agent goes in the wrong direction, send a short, clear steer message instead of stopping and restarting the turn; the existing context is preserved and the work is corrected in the same turn.

  • APIKimi Code server API

    The session list’s `busy` filter is applied before pagination

    `GET /api/v1/sessions?busy=true` now correctly finds busy sessions that would otherwise fall behind new and idle sessions on the page. `has_more` is calculated based on filtered results, and page size is no longer consumed by non-busy records.

    Usage details

    Who can use it: On the official kimi-code repository’s main branch; not listed separately in the release notes for 2.5.3, released on November 10. Check whether it is available in your installation with `kimi --version` and in subsequent release notes. Kimi Code server API session-listing endpoints and busy/idle session panels.

    How to use it: Use `GET /api/v1/sessions?busy=true&page_size=1`; evaluate the response’s `items` and `has_more` fields based only on busy results.

    Tip: If a filtered page is empty, the client should also check the `has_more` flow before immediately concluding that there are no busy sessions.

  • APIKimi Code server API

    `agent_config` and custom metadata are applied when creating a session

    `POST /api/v1/sessions` now applies submitted partial `agent_config` values before returning the session and stores custom metadata keys. `tower_mode: true` is rejected during creation; Tower should be enabled through the profile endpoint after the session is created.

    Usage details

    Who can use it: On the official kimi-code repository’s main branch; not listed separately in the release notes for 2.5.3, released on November 10. Check whether it is available in your installation with `kimi --version` and in subsequent release notes. Web, Desktop, and custom clients that create sessions through the Kimi Code server API.

    How to use it: In the `POST /api/v1/sessions` body, send custom keys in `metadata` and the same valid fields as in the profile in `agent_config`; for Tower, use `POST /api/v1/sessions/{session_id}/profile` afterward.

    Tip: Don’t include `tower_mode: true` in the creation request; set it after session setup to avoid a failed repository change.

  • CapabilityKimi Code server API

    Running tool calls interrupted by Stop are marked `interrupted`

    When the user stops a turn, tool calls that are still running now close with an `interrupted` outcome instead of being marked as failed (`failed`). The results of calls completed before the stop and calls that encountered actual errors are preserved.

    Usage details

    Who can use it: On the main branch of the official kimi-code repository; not separately listed in the 2.5.3 release notes published on November 10. Check whether it is available in your installation with `kimi --version` and subsequent release notes. Kimi Code Web conversation history, transcript, and server API clients.

    How to use it: Use Stop during a running turn; handle the `tool_call.outcome` field in the history separately as `interrupted`, `failed`, or `completed`.

    Tip: Before converting `interrupted` into a retry step, treat the user's Stop action as an intentional cancellation.

  • CapabilityKimi Code

    Preview and download headers for Unicode filenames fixed

    Download responses for filenames containing non-ASCII characters now include a safe ASCII fallback name along with UTF-8 `filename*` information. This prevents Chinese, Turkish, and other Unicode filenames from being corrupted in file previews and downloads.

    Usage details

    Who can use it: On the main branch of the official kimi-code repository; not listed separately in the release notes for 2.5.3, released on November 10. Check whether your installation includes it with `kimi --version` and subsequent release notes. For Kimi Code server file download endpoints and local or remote workspaces using Unicode filenames.

    How to use it: In Kimi Code web/TUI, select a file with a Unicode name using `@`, or use the session `/api/v1/sessions/{session_id}/fs/{path}:download` endpoint; URL-encode the path.

    Tip: On the client side, read `filename*` first, and use the ASCII `filename` fallback if it is unavailable.

  • Plugin / IntegrationKimi Code

    stdio MCP scripts such as `npx` now start on Windows

    Starting stdio MCP servers defined with batch/script commands such as `npx` on Windows has been fixed. Persistent process errors similar to `ENOENT` during startup are now also classified more accurately as configuration errors.

    Usage details

    Who can use it: Kimi Code 2.5.3 (released on npm on November 10) and later. Kimi Code on Windows and MCP servers using stdio transport.

    How to use it: In the MCP configuration, set the transport type to `stdio`, specify `npx` or a similar script as the command, and define its arguments, then restart Kimi Code on Windows.

    Tip: First run the MCP command in a terminal under the same Windows account to verify PATH access; if the error persists, check the configuration classification.

  • CapabilityKimi Code server API

    Unicode filenames of uploaded files are preserved on download

    Download responses for uploaded files now preserve Unicode filenames in UTF-8 using `filename*`; non-ASCII characters are no longer lost by being converted to underscores. The change particularly covers upload-based file downloads such as `GET /api/v1/files/{file_id}`.

    Usage details

    Who can use it: On the main branch of the official kimi-code repository; not listed separately in the release notes for 2.5.3, released on November 10. Check whether your installation includes it with `kimi --version` and subsequent release notes. For clients that upload files to the Kimi Code server API and download them later.

    How to use it: After uploading a file, read the `Content-Disposition` header in the `GET /api/v1/files/{file_id}` response; decode the Unicode name from `filename*`.

    Tip: If your download library reads only the legacy `filename` field, update it to display UTF-8 filenames correctly.

  • FeatureKimi Code server API

    Cron messages show only the scheduled prompt

    In the conversation view, user messages triggered by Cron now show only the scheduled prompt text, instead of the internal `<cron-fire>` wrapper and scheduling metadata. Older and nested wrapper formats are also safely unwrapped.

    Usage details

    Who can use it: On the official kimi-code repository’s main branch; not listed separately in the release notes for 2.5.3, released on November 10. Check whether it is available in your installation with `kimi --version` and in subsequent release notes. Kimi Code cron/scheduled tasks and Web/API clients that display their conversation history.

    How to use it: Wait for a cron task to run, then open the user message in the conversation history; the actual prompt text appears directly, without wrapper information mixed into the user text.

    Tip: If the client parses cron envelopes on its end, display the text shown in the new history directly without extracting it again.

  • CapabilityKimi Code server API

    Additional `note` information on tool results is preserved in conversation items

    Explanatory information added to a tool result—for example, file size and image dimensions for `ReadMediaFile`—is now preserved alongside the output in the `tool_call.note` field. This lets the client display this supplementary information separately from the output text.

    Usage details

    Who can use it: On the main branch of the official kimi-code repository; not separately listed in the 2.5.3 release notes published on November 10. Check whether it is available in your installation with `kimi --version` and subsequent release notes. Clients consuming Kimi Code server API conversation history and WebSocket `tool_call` items.

    How to use it: When processing conversation items, read the optional `tool_call.note` field alongside `tool_call.output` and display it in a media or file information panel.

    Tip: Do not interpret the `note` field as model output; present it separately as supplementary metadata that can be shortened.

  • CapabilityKimi Code

    The Skill catalog remains visible even when the Skill tool is disabled

    Even when the Skill tool is disabled, available skill names, descriptions, and paths remain in the system prompt catalog; only tool invocation is blocked. Skill change notifications also continue to appear in this state.

    Usage details

    Who can use it: On the main branch of the official kimi-code repository; not listed separately in the release notes for 2.5.3, released on November 10. Check whether your installation includes it with `kimi --version` and subsequent release notes. For Kimi Code workspaces and profiles where the Skill tool is disabled by session, profile, or global tool policy.

    How to use it: Disable the `Skill` tool in the agent profile or session tool policy; the catalog remains in the model’s system prompt under the `# Skills`/`${skills}` section.

    Tip: Keeping the catalog visible in profiles where you want to restrict Skill calls lets the model know which skills are available.

  • AvailabilityAmazon Bedrock

    Kimi K3.1 Pro Highspeed now available on Amazon Bedrock

    The high-output-speed Highspeed version of Kimi K3.1 Pro has arrived on Amazon Bedrock; according to AWS, Moonshot says this version is 2–3 times faster than Kimi K3.1 Pro. The model supports tool calling, structured output, adjustable reasoning, visual input, and streaming; data is processed within AWS boundaries, not shared with Moonshot, and zero data retention is always enabled for inference. The Flex tier is available at half the Standard price.

    Usage details

    Who can use it: With cross-Region inference in all AWS regions where Amazon Bedrock is available: `global.moonshotai.kimi-k3.1`, US-only `us.moonshotai.kimi-k3.1`, India-only `in.moonshotai.kimi-k3.1`. The AWS announcement is dated November 6; the detailed blog is dated November 10.

    How to use it: To try it without writing code, select Kimi K3.1 from Test > Playground in the Bedrock console. In code, point the OpenAI SDK to `https://bedrock-runtime.<region>.amazonaws.com/openai/v1` and set the model to `global.moonshotai.kimi-k3.1`; set `reasoning_effort` to `low`, `high`, or `max` (default `max`). For low-cost background tasks, add `service_tier: "flex"` to the request.

    Tip: For simple tasks, reduce latency and output tokens with `reasoning_effort: "low"`; move non-urgent tasks such as evaluations and batch summarization to the Flex tier to cut costs in half.

Tencent · Hy

Tencent Hy4
Full history (369) →

New in this edition · 1

  • OtherGroupAssistBench

    Evaluation data for the GroupAssistBench dataset updated

    The Tencent Yuanbao team updated the GroupAssistBench data package with a November 9 `update data` commit. The changes cover nine files, including evaluation inputs, conversation contexts, cut points, a manifest, and profile files; the card provides 2,395 test examples from 50 group chats.

    Usage details

    Who can use it: Publicly available on Hugging Face; includes 2,395 test examples, with English reference translations alongside the Chinese data. The dataset is available for non-commercial use only under CC BY-NC 4.0.

    How to use it: Load the test split in Python with `load_dataset('tencent/GroupAssistBench', split='test')`. Download the conversation history with `hf_hub_download('tencent/GroupAssistBench', 'dataset/contexts.json', repo_type='dataset')`, or get the full package with `hf download tencent/GroupAssistBench --repo-type dataset --local-dir GroupAssistBench`.

    Tip: Report L1 memory recall and L2 memory use results separately by M1/M2/M3 memory dependencies; for each query, provide the model only with the history before `cut_idx`.

Alibaba · Qwen

Qwen3.8 Max
Full history (673) →

New in this edition · 37

  • AgentQwen Code Managed Agent

    Private BTW follow-ups and session recap added to Managed Agent

    Managed Agent now supports private BTW follow-ups and manual or optionally automatic session recaps. Automatic recap is off by default; Web Shell panels use persistent results.

    Usage details

    Who can use it: Qwen Code Managed Agent and Web Shell panels.

    How to use it: Use private `/btw` follow-ups in a Managed Agent session; manually run the recap action in the session panel or enable automatic recap.

    Tip: Enable automatic recap only for long sessions; keep the default off setting for short tasks to reduce unnecessary recap calls.

    ✓ Verified by official sourceSource: Qwen Code v0.27.2 sürüm notları
  • APIQwen Code CLI

    Qwen Code CLI v0.27.2 released

    Qwen Code CLI v0.27.2 has been released. The release includes user-facing changes such as new channel management, Mods capabilities, Managed Agent summaries, and approval notifications.

    Usage details

    Who can use it: Qwen Code CLI users; installations upgrading from v0.27.1.

    How to use it: Install the v0.27.2 package from the official GitHub Releases page or update your existing Qwen Code installation; then verify the version with `qwen --version`.

    Tip: Test channel and Mods configurations in a separate workspace with v0.27.2 before using them in production automation.

    ✓ Verified by official sourceSource: Qwen Code v0.27.2 sürüm notları
  • APIQwen Code TypeScript SDK

    TypeScript SDK v0.1.32 bundles CLI v0.27.2

    Qwen Code TypeScript SDK v0.1.32 has been released and bundles CLI v0.27.2. This is the new SDK release after the known v0.1.31/CLI 0.27.1 entry.

    Usage details

    Who can use it: Developers using the Qwen Code SDK with Node.js.

    How to use it: Run `npm install @qwen-code/sdk@0.1.32` in your project and verify the CLI version included in the package.

    Tip: Use the bundled CLI instead of installing SDK and CLI versions separately to reduce the risk of version incompatibility.

  • Plugin / IntegrationQwen Code Web Shell

    QR sign-in WeChat channel added to Web Shell

    The WeChat channel can be configured through Web Shell Channel Manager. In-browser QR sign-in supports scan progress, identity display, sign-out, and refreshing expired codes.

    Usage details

    Who can use it: Qwen Code Web Shell users; workspaces connecting a WeChat account as a channel.

    How to use it: Open Channel Manager in Web Shell, add the WeChat channel, scan the QR code, and save the channel after authentication is complete.

    Tip: If the QR code expires, use the refresh action; check the channel identity before signing out.

    ✓ Verified by official sourceSource: Qwen Code v0.27.2 sürüm notları
  • FeatureQwen Code CLI

    `/btw` side questions are replayed in session context

    `/btw` side questions are now part of the session, and the last 20 answered questions are replayed in follow-up questions. This lets you continue based on previous answers to side questions.

    Usage details

    Who can use it: Qwen Code CLI/TUI sessions.

    How to use it: Ask a side question with `/btw` in a session, then send a follow-up question in the same session; the last answered side questions are automatically added to the context.

    Tip: Keep decision points in `/btw` without cluttering the main task; be mindful of the 20-record limit in follow-up questions.

    ✓ Verified by official sourceSource: Qwen Code v0.27.2 sürüm notları
  • AvailabilityQwen Code Managed Agent

    Tool v4 execution metrics added to Activity diagnostics

    Existing Managed Agent Activity diagnostics can now show Tool v4 execution metrics. Duration, valid sample count, and proposed versus applied line changes are broken out.

    Usage details

    Who can use it: Qwen Code Managed Agent Activity/diagnostics view; new metrics coverage for existing execution diagnostics.

    How to use it: Open the Activity or diagnostics view in a Managed Agent session and review duration and the proposed/applied line changes fields in the Tool v4 metric rows.

    Tip: Use the difference between proposed and applied lines to assess the proportion of changes the model actually applied.

    ✓ Verified by official sourceSource: Qwen Code v0.27.2 sürüm notları
  • SettingQwen Code CLI

    Pending approvals appear in the live panel and trigger terminal/desktop alerts

    Background and workflow approvals are now visible in the live panel; a terminal bell or desktop notification can be triggered when the terminal is not focused. This behavior follows the `general.terminalBell` and `general.notificationMode` settings.

    Usage details

    Who can use it: Qwen Code CLI/TUI; unfocused terminals and supported desktop notifications.

    How to use it: Configure `general.terminalBell` and `general.notificationMode` in Settings; check the live panel while a background agent is waiting for approval.

    Tip: Use separate profiles to enable desktop notifications for long-running tasks and disable the terminal bell in quiet CI environments.

    ✓ Verified by official sourceSource: Qwen Code v0.27.2 sürüm notları
  • AvailabilityQwen Code Web Shell

    Email Channel can be configured from Channel Manager

    A direct configuration interface has been added to Web Shell Channel Manager for existing Email Channel support. This is the new management and setup flow for the established remote-task email channel.

    Usage details

    Who can use it: Workspaces using Email Channel in Qwen Code Web Shell.

    How to use it: Open Web Shell → Channel Manager → Email, fill in the required channel fields, and save the connection.

    Tip: Before putting the email channel into production, send a small test task and check delivery and the response loop.

    ✓ Verified by official sourceSource: Qwen Code v0.27.2 sürüm notları
  • AvailabilityQwen Code GitHub Channel

    PR events in the GitHub channel can be restricted to an active watch

    The `requireBoundWatch` option has been added to the GitHub channel. Review or verify bindings can receive PR events only while an `/autofix-pr` or `/verify-pr` watch is active.

    Usage details

    Who can use it: Workspaces using PR review/verify bindings with GitHub Channel.

    How to use it: Enable `requireBoundWatch` in the GitHub Channel configuration; then start an `/autofix-pr` or `/verify-pr` watch on the relevant PR.

    Tip: Enable this option to reduce unwanted PR deliveries, and explicitly manage the watch lifecycle in your automation.

    ✓ Verified by official sourceSource: Qwen Code v0.27.2 sürüm notları
  • CapabilityQwen Code GitHub Channel

    GitHub PR watch saves quota on unchanged responses

    The GitHub channel’s PR status monitoring and notification lists use conditional requests and caching. API quota is preserved by reducing repeated data consumption for unchanged responses.

    Usage details

    Who can use it: Workspaces using GitHub Channel PR watch and notification lists.

    How to use it: Enable GitHub Channel PR watch as usual; v0.27.2 automatically handles unchanged responses with conditional requests.

    Tip: Use this behavior in workspaces that require frequent polling; continue monitoring GitHub’s main rate limit nonetheless.

    ✓ Verified by official sourceSource: Qwen Code v0.27.2 sürüm notları
  • APIQwen Code Channel API

    The channel delivery resource returns a segments array

    A `segments` array has been added to the public channel delivery resource. Legacy single-segment fields are retained for compatibility.

    Usage details

    Who can use it: API clients that read the channel delivery resource and multipart deliveries.

    How to use it: Add client logic that uses the new `segments` array when reading the API response; retain legacy single fields as a fallback during the transition.

    Tip: For multipart messages, match delivery receipts using the `segments` array rather than the single field.

    ✓ Verified by official sourceSource: Qwen Code v0.27.2 sürüm notları
  • SettingQwen Code Mods

    Mods can persist values across sessions with $.store

    Each plugin’s values can now be persisted across sessions with `$.store`. This preserves Mod configurations after a restart.

    Usage details

    Who can use it: Qwen Code Mods plugins.

    How to use it: Write and read persistent values through `$.store` in your mod code; keep plugin-specific keys in a separate namespace.

    Tip: Don’t write temporary results to the persistent store; save only user settings or small state values.

    ✓ Verified by official sourceSource: Qwen Code v0.27.2 sürüm notları
  • FeatureQwen Code Mods

    Mods can call one-shot model completion

    Mods can now run one-shot text completion with `$.model.complete(request, {signal})`. The call uses the session client without tools, history, or an additional system prompt.

    Usage details

    Who can use it: Qwen Code Mods plugins.

    How to use it: Use `$.model.complete(request, {signal})` in your mod code and connect the returned completion to your plugin’s flow.

    Tip: Use this API for standalone short generations; prefer the main agent flow if you need session history or tool context.

    ✓ Verified by official sourceSource: Qwen Code v0.27.2 sürüm notları
  • FeatureQwen Code Mods

    Mods can replace Ink response lines with ui.render

    Mods can rewrite the model response text shown in the Ink terminal through the `AssistantMessage` render point and the `ui.render` hook. Saved or copied text remains unchanged.

    Usage details

    Who can use it: Ink terminal plugins using Qwen Code Mods.

    How to use it: Define the `ui.render` hook in your mod, transform the displayed AssistantMessage text, and keep the same response separate from the saved content.

    Tip: Change only the presentation layer; preserve the original record for user oversight and debugging.

    ✓ Verified by official sourceSource: Qwen Code v0.27.2 sürüm notları
  • AvailabilityQwen Code Web Shell

    QQ channels can be managed from Web Shell with QR codes

    QQ channel configuration and optional QR sign-in management have been added to Web Shell. This lets you connect and manage an existing QQ bot from Web Shell by scanning the account’s QR code.

    Usage details

    Who can use it: Users with a QQ bot in Qwen Code Web Shell.

    How to use it: Open the QQ channel in Channel Manager, enable QR sign-in management, scan the code, then verify and save the connected identity.

    Tip: With multiple QQ accounts, check the displayed identity before saving to prevent misdelivery.

    ✓ Verified by official sourceSource: Qwen Code v0.27.2 sürüm notları
  • SettingQwen Code Managed Agent

    The self-hosted autofix agent budget increased to 240 minutes

    Self-hosted autofix runs now use a 240-minute agent budget. This reduces the risk of early timeouts during long autofix runs.

    Usage details

    Who can use it: Self-hosted `/autofix-pr` runs.

    How to use it: Start a self-hosted autofix run as usual; the new agent budget is automatically set to 240 minutes.

    Tip: Use the time budget for long build and test chains, but configure PR monitoring conditions separately to limit failed loops.

    ✓ Verified by official sourceSource: Qwen Code v0.27.2 sürüm notları
  • SettingQwen Code Channel API

    The authorExclude field is format-validated before delivery

    Invalid `authorExclude` values in incoming task records are now rejected before reaching comment delivery. This prevents crashes or invalid delivery behavior caused by misconfiguration.

    Usage details

    Who can use it: GitHub and connected channel flows that use `authorExclude` in inbound task/channel records.

    How to use it: Correct the `authorExclude` values in your inbound task configuration to match the expected format; starting with v0.27.2, records are validated before delivery.

    Tip: Validate authorExclude with a single sample inbound record before enabling the channel.

    ✓ Verified by official sourceSource: Qwen Code v0.27.2 sürüm notları
  • FeatureQwen Code CLI

    Idle-only slash commands work in empty Ink sessions

    Idle-only commands such as `/branch`, `/cd`, `/fork`, `/recap`, and `/advisor` are no longer blocked by application state in an idle Ink TUI session.

    Usage details

    Who can use it: Idle Qwen Code CLI sessions in Ink TUI.

    How to use it: While the Ink TUI session is idle, enter the relevant slash command; for example, run `/branch` or `/recap` directly.

    Tip: Refresh the context with `/recap` before starting a new turn; don’t expect the command to be rejected in an empty session.

    ✓ Verified by official sourceSource: Qwen Code v0.27.2 sürüm notları
  • CapabilityQwen Code GitHub Channel

    gh pr create context is bound to the working directory

    Pull requests created with `gh pr create` are now bound to the directory where the command is run. This makes worktree and PR session matching more accurate.

    Usage details

    Who can use it: Qwen Code sessions linked to GitHub PRs and developers using worktrees.

    How to use it: Switch to the correct worktree directory and run `gh pr create`; Qwen Code establishes the PR link from the command’s current working directory.

    Tip: When creating a PR, use `pwd` to check that you are in the target worktree directory rather than the main repository.

    ✓ Verified by official sourceSource: Qwen Code v0.27.2 sürüm notları
  • CapabilityQwen Code CLI

    Goal verifier displays the Last check explanation in the session language

    The Last check reason shown in the Goal verifier decision now follows the session’s output language. The decision values, accept and reject, remain unchanged.

    Usage details

    Who can use it: CLI/TUI sessions using Qwen Code Goal verification.

    How to use it: Run a Goal and inspect the Last check reason in the session output; continue using accept/reject values in the decision field.

    Tip: Show the localized explanation to users, and rely on the fixed accept/reject fields when parsing automation output.

    ✓ Verified by official sourceSource: Qwen Code v0.27.2 sürüm notları
  • FeatureQwen Code Web Shell

    A readable data table is shown when a chart cannot be generated

    When the chart generator fails and inline data is readable, Web Shell now falls back to the Data table view. Instead of seeing that the chart is unavailable, users can continue examining the data.

    Usage details

    Who can use it: Agent outputs containing charts or tables in Web Shell.

    How to use it: Generate an output containing a chart; if chart generation fails, open the Data table view in the same card.

    Tip: Check important decisions against the fallback data table as well as the chart.

    ✓ Verified by official sourceSource: Qwen Code v0.27.2 sürüm notları
  • SettingQwen Code Channel API

    Channel fields support an inclusive lower bound for zero values

    Channel configuration fields can now define an inclusive lower bound. This ensures fields where zero is valid are treated correctly in validation and range messages.

    Usage details

    Who can use it: Channel configuration schemas and Web Shell/CLI clients that use them.

    How to use it: Define an inclusive lower bound in the channel field schema and revalidate a configuration with a zero value.

    Tip: For numeric settings where zero is meaningful, explicitly use an inclusive lower bound to avoid differences in client validation.

    ✓ Verified by official sourceSource: Qwen Code v0.27.2 sürüm notları
  • AvailabilityQwen Code Agent Host

    A lost enrollment response triggers an Agent Host retry

    Agent Host can safely retry a lost enrollment or Replace response. The existing Host is completed, reducing the risk of unnecessary errors or duplicate Host creation.

    Usage details

    Who can use it: Managed Agent Host enrollment/Replace flows.

    How to use it: If an enrollment or Replace request response is lost, resend the same operation; refresh the Host status and check that no duplicate registration was created.

    Tip: After a network outage, retry enrollment once before creating a new Host.

    ✓ Verified by official sourceSource: Qwen Code v0.27.2 sürüm notları
  • CapabilityQwen Code CLI

    Error for unavailable saved subscription model deferred until submission

    If the saved subscription model is no longer available, the CLI opens the composer silently and defers the error until submission. Users can see the gateway response and resolve the issue.

    Usage details

    Who can use it: CLI sessions using a Qwen subscription model.

    How to use it: Select the saved model, open the composer, and submit the request; if the model is unavailable, review the gateway error in the submission response.

    Tip: Check the model or access information in the gateway response before selecting the model again.

    ✓ Verified by official sourceSource: Qwen Code v0.27.2 sürüm notları
  • SettingQwen Code Web Shell

    Managed sessions can reuse the host shell header

    Managed Web Shell sessions can reuse the host shell chrome with the `showManagedSessionHeader` and `alignManagedChatWithHost` options. This reduces duplicate headers and hints.

    Usage details

    Who can use it: Setups that use Managed Web Shell sessions inside an embedded host shell.

    How to use it: Enable the `showManagedSessionHeader` and `alignManagedChatWithHost` options in the Web Shell managed session configuration, then check the header alignment.

    Tip: Try both options together to save space in the embedded view; keep the existing header in standalone sessions.

    ✓ Verified by official sourceSource: Qwen Code v0.27.2 sürüm notları
  • AvailabilityQwen Code OpenTUI

    OpenTUI model and tool turns moved to the unified agent loop

    With `experimental.unifiedAgentLoop` enabled, OpenTUI model turns and tool batches now run through the shared agent loop. This brings OpenTUI behavior more in line with Ink.

    Usage details

    Who can use it: OpenTUI users who use the experimental unifiedAgentLoop.

    How to use it: Enable `experimental.unifiedAgentLoop` in the settings configuration, start OpenTUI, and verify that model and tool turns go through the shared loop.

    Tip: Before using the experimental flow in production, compare it with its Ink equivalent using the same prompt and tool set.

    ✓ Verified by official sourceSource: Qwen Code v0.27.2 sürüm notları
  • AvailabilityQwen Code CLI

    Automatic session titles use the language of the conversation

    Automatically generated session titles are now written in the language of the conversation. This adds language adaptation to the behavior that preserves known titles when reconstructing transcripts.

    Usage details

    Who can use it: CLI/TUI and Web Shell flows that use automatic titles in Qwen Code sessions.

    How to use it: Start a new session and wait for an automatic title to be generated; the title follows the selected language of the conversation.

    Tip: Check automatic titles before renaming them manually to maintain language consistency in session searches.

    ✓ Verified by official sourceSource: Qwen Code v0.27.2 sürüm notları
  • SettingQwen Code CLI

    Sandbox debugging is enabled only by explicit DEBUG values

    The CLI now enables sandbox debugging only when DEBUG values are explicitly specified. Ambiguous or accidentally passed truthy values do not enable debug mode.

    Usage details

    Who can use it: Qwen Code CLI developers using sandboxing.

    How to use it: When debugging is needed, set a supported explicit `DEBUG` value in the environment and start the CLI; remove the value to return to normal behavior.

    Tip: Keep the DEBUG environment variable clean in production shells; enable verbose sandbox logs only while debugging.

    ✓ Verified by official sourceSource: Qwen Code v0.27.2 sürüm notları
  • CapabilityQwen Code CLI

    Shell permission analysis recognizes ANSI-C escapes and the -c option terminator

    Shell command parsing now handles ANSI-C escapes and `-c` option terminators correctly. Self-kill protection and file permission checks work more accurately with these command forms.

    Usage details

    Who can use it: CLI sessions where Qwen Code analyzes Bash/shell commands for permissions and security.

    How to use it: Run commands that use ANSI-C quoting or a `-c` terminator as usual; v0.27.2 applies the parsing automatically.

    Tip: Before using complex shell commands in production, check the permission decision with a dry run or in a low-privilege workspace.

    ✓ Verified by official sourceSource: Qwen Code v0.27.2 sürüm notları
  • FeatureQwen Code Web Shell

    Web Shell slash menu shows command names first

    The Web Shell slash menu now truncates descriptions first, prioritizing command and Skill names. Full details are available on hover.

    Usage details

    Who can use it: Qwen Code Web Shell slash command and Skill menu.

    How to use it: Type `/` in the Composer to open the slash menu; view command names and hover over a row for details.

    Tip: For long Skill descriptions, type the beginning of the name to find it faster than searching by description.

    ✓ Verified by official sourceSource: Qwen Code v0.27.2 sürüm notları
  • SettingHiRoute

    Reasoning effort can be selected in Codex fixed plans

    Reasoning effort selection has been added for Codex runs tied to fixed routing plans. When the plan is saved again and Codex is restarted, each candidate model uses its configured effort value.

    Usage details

    Who can use it: Codex users who use a HiRoute routing plan.

    How to use it: Open the reasoning selector for the fixed plan in Codex, choose an effort level, save the model routing again, and restart Codex.

    Tip: Set the plan’s cost and quality balance by choosing low effort for simple candidate models and higher effort for critical candidates.

    ✓ Verified by official sourceSource: HiRoute 0.6.0 sürüm notları
  • AvailabilityHiRoute

    Signed Omarchy desktop packages released for HiRoute

    Signed desktop packages for ARM64 and x86_64 on Omarchy were released with HiRoute 0.6.0. The package includes a launcher, desktop application, and verifiable standalone workload.

    Usage details

    Who can use it: Omarchy/Arch Linux users; an additional platform alongside the currently supported macOS, Windows, and Linux packages.

    How to use it: Download the HiRoute 0.6.0 package, .sig/.sha256 files, and release key; verify the SHA-256 and GPG fingerprint, then install with pacman.

    Tip: Keep the package, signature, and checksum files together; verify the CB5F8ADEDC272ED8879D206C5642FA57EA8DF169 fingerprint before installing.

  • FeatureHiRoute

    Route and publication references can be previewed before deleting a decision model

    Active route, draft, and publication references can now be previewed before removing a decision model. Deletion preserves usage history, shared API connections, and keys.

    Usage details

    Who can use it: HiRoute Decision Models administrators.

    How to use it: In the Decision Models section, select the model to remove; review and confirm the active route, draft, and publication references shown before deletion.

    Tip: Move active routes to another decision model before deleting; preserving history and shared keys makes recovery easier.

    ✓ Verified by official sourceSource: HiRoute 0.6.0 sürüm notları
  • Pricing / PlanHiRoute

    Shared decision calls included in API key allowance

    Shared calls from decision models are now reserved against the API key allowance and reconciled in the final total. Calls with unknown prices can still run but remain unpriced.

    Usage details

    Who can use it: HiRoute connections and decision models using an API key allowance.

    How to use it: Set the API key allowance in the HiRoute connection; run the decision model with the same key and track reservation and reconciliation records.

    Tip: Set the budget to cover not only regular model calls but also the portion allocated to decision model calls.

    ✓ Verified by official sourceSource: HiRoute 0.6.0 sürüm notları
  • CapabilityHiRoute

    Responses sessions preserve tool and reasoning continuity after compaction

    In Responses sessions routed through Codex, tool calls and reasoning remain linked across compaction and subsequent requests. Portable reasoning summaries are preserved even when the model changes, and scoring evidence from long tool sessions is not lost.

    Usage details

    Who can use it: Users of Codex/Responses with HiRoute; long and compacted sessions.

    How to use it: Run the Codex connection through HiRoute; continue the same session after compaction or a continuation request in a long Responses session.

    Tip: For long tasks, use continuation after compaction instead of splitting the session history to preserve prior evaluation evidence.

  • CapabilityHiRoute

    Hermes preserves provider routing attribution in resumed sessions

    When a resumed Hermes session selects a named provider, HiRoute routing attribution is preserved. This ensures the selected provider is not lost from the session history.

    Usage details

    Who can use it: Users resuming a HiRoute session with Hermes.

    How to use it: Resume the Hermes session through the HiRoute plan and check that the named provider selection remains in the session attribution during resumption.

    Tip: When troubleshooting, keep provider attribution as well as the model name; this makes it easier to distinguish the wrong connection in a resumed session.

    ✓ Verified by official sourceSource: HiRoute 0.6.0 sürüm notları
  • CapabilityHiRoute

    Windows Scheduler ownership checks and Gateway replay recovery improved

    HiRoute improved Windows Task Scheduler ownership checks and Gateway replay recovery after temporary working directories are cleaned up. This enables more reliable continuation after service restarts and temporary file cleanup.

    Usage details

    Who can use it: Installations using the HiRoute service and Task Scheduler on Windows.

    How to use it: Run the HiRoute service with Windows Scheduler; after the temporary working directory is cleaned up, restart the Gateway replay flow and verify automatic recovery.

    Tip: If you encounter an issue, save the service logs before manually deleting the temporary directory; you can see at which stage replay recovery took effect.

    ✓ Verified by official sourceSource: HiRoute 0.6.0 sürüm notları

Watchlist

Muse Wonder

44
Meta · Meta AI

Meta’s standard Muse Wonder model; 11th with 44 points. Available in preview on Google Agent Platform as `muse-wonder-1.0`.

Claude Haiku 5.5

36
Anthropic · Claude

Anthropic’s fast, low-cost model at $0.10 / $0.50. Meta says its new Muse Ember delivers the same score at about one-third the task cost.

Muse Ember

36
Meta · Meta AI

Released on November 9. Meta’s smallest Muse model; $0.10 / $0.50, 1M context, available as `muse-ember-1.0` in Meta Model API and on all Muse Code plans.

Grok 4.7

35
xAI · Grok

The default model in the Grok app; text outputs in Grok API now have a watermark (`x-watermark`) by default. No new changes at xAI since November 9.

Tencent Hy4

35
Tencent · Hy

Available via API on OpenRouter (`tencent/hy4`); 50% off on TokenHub through December 31, free on WorkBuddy and CodeBuddy through November 15.

Qwen3.8 Max

34
Alibaba · Qwen

Alibaba’s flagship cloud model; Qwen 4 has not launched yet. Qwen Code 0.27.2 brought WeChat and QQ channels that connect to Web Shell via QR code.

Sources