Gemini 4 Carbon takes the lead: just a 2-point gap among the top three models

New in this edition · 7

• 30 Oct
  APIGemini API

  Computer Use tool reaches general availability

  The Computer Use tool is now generally available (GA) on the Interactions API and GenerateContent API. Developers can now use Gemini-based systems in production to take actions in browser, mobile, and desktop interfaces.

  Usage details

  Who can use it: Gemini API users; projects using the Interactions API and GenerateContent API.

  How to use it: Add the Computer Use tool from the Gemini API documentation to your request’s tool configuration.

  Tip: Before deploying to production, restrict action permissions and steps requiring human approval with separate security policies.

  ✓ Verified by official sourceModels: Gemini 3.8 Flash, Computer UseSource: Release notes | Gemini API - Google AI for Developers
• 29 Oct
  FeatureGemini app

  Files can be created and edited in Gemini chats

  Gemini can now create, preview, and edit documents, PDFs, presentations, spreadsheets, Markdown, and code files in chat. PDF editing is supported on web and mobile; other formats are supported on web, and results can be exported to Google Slides, Sheets, or Drive.

  Usage details

  Who can use it: Personal accounts with access to the Gemini app, Workspace Individual, and Google Workspace users; in all supported languages.

  How to use it: Ask Gemini to create the file type you want in a chat; open the preview full-screen, select text, and request edits.

  Tip: Check the structure and sources in the Gemini preview before exporting a presentation or spreadsheet.

  ✓ Verified by official sourceSource: Create, preview, and edit files directly in the Gemini app
• 28 Oct
  ModelGemini models

  Gemini 4 Carbon announced

  Google announced Gemini 4 Carbon, focused on long-running coding and agent workflows, as well as cybersecurity defense. The model is initially available to a limited group of cyber defenders through the Fairwind Program; general access for developers, businesses, and consumers is planned for later.

  Usage details

  Who can use it: For now, selected cyber defenders and Fairwind Program partners; general user and developer access is not yet available.

  How to use it: Once general access is available, use the Gemini 4 Carbon model selection in the Gemini API, product surfaces, or developer consoles.

  Tip: For cybersecurity use, add independent security testing and human approval before automatically applying model outputs.

  ✓ Verified by official sourceModels: Gemini 4 CarbonSource: Gemini 4 Carbon: our next frontier model
• 28 Oct
  CapabilityGemini in Google Sheets

  Fill with Gemini now shows sources at the cell level

  In Google Sheets, Fill with Gemini now shows the web sources it uses separately in each cell. Request capacity has increased to 3,000 cells, while processing speed and consistency across cells have also improved.

  Usage details

  Who can use it: Users with Gemini in Sheets enabled on Business Standard/Plus, Enterprise Standard/Plus, Google AI Pro/Ultra, Google AI Pro for Education, and AI Expanded Access.

  How to use it: In Google Sheets, open the Ask Gemini icon in the upper right and fill cells with AI function; sources appear in the cells.

  Tip: Before using generated data, verify the source links in cells by sampling them.

  ✓ Verified by official sourceSource: Fill with Gemini now provides cell-level citations for web sources and supports larger requests

+3 →

New in this edition · 10

• 30 Oct
  SettingClaude Code

  Permission setting blocks writes outside the working directory

  The new permissions.blockWritesOutsideWorkingDirectories setting makes the Write, Edit, and NotebookEdit tools reject paths outside working directories in every permission mode. This provides an additional security layer that narrows file access for sessions running automatically.

  Usage details

  Who can use it: Claude Code 2.1.314 and later.

  How to use it: Enable permissions.blockWritesOutsideWorkingDirectories in managed or user settings.

  Tip: Explicitly define working directories and enforce this setting in CI and on shared machines.

  ✓ Verified by official sourceSource: Claude Code 2.1.314 sürüm notları
• 30 Oct
  FeatureClaude Code

  Automatic idle compaction comes to Claude Code

  Conversations exceeding 200,000 tokens and left idle in VS Code and the Claude Desktop Code tab are automatically summarized. The summary is used in the message after the prompt cache expires, making it easier to continue long sessions.

  Usage details

  Who can use it: For users of Claude Code in VS Code and the Claude Desktop Code tab; coverage has expanded subject to provider and telemetry conditions.

  How to use it: Update Claude Code; leave the Idle compaction option in /config enabled in settings.

  Tip: Before pausing a long task, check context usage with /context and /usage.

  ✓ Verified by official sourceSource: Claude Code changelog
• 29 Oct
  APIClaude API/Developer Platform

  Usage and Cost Analytics API launches in beta

  The new Usage and Cost Analytics API offers two beta endpoints for reporting token usage and costs for Claude Console and Claude Enterprise organizations. Data can be filtered and grouped by dimensions such as model, workspace, API key, and user or service account.

  Usage details

  Who can use it: Beta for Claude Console and Claude Enterprise organizations; the analytics-usage-cost-2026-08-24 beta header is required.

  How to use it: Send POST requests to the /v1/organizations/analytics/usage and /v1/organizations/analytics/cost endpoints with the relevant beta header.

  Tip: Track costs by workspace and service account as well as by model to identify budget variances.

  ✓ Verified by official sourceSource: Claude Platform API sürüm notları
• 29 Oct
  RetiredClaude API/Developer Platform

  Legacy Usage and Cost reports are now legacy

  The usage_report, cost_report, user_usage_report, and user_cost_report endpoints in the Usage and Cost API and Enterprise Analytics API have been marked legacy. Existing integrations will continue to work, but migration to the Analytics API is recommended for new development.

  Usage details

  Who can use it: Support continues for existing users; the new Analytics API is recommended for new integrations.

  How to use it: Follow the Migrate from the Usage and Cost API guide to move to the new usage and cost endpoints.

  Tip: Instead of removing the old endpoints immediately, compare their results with the new API for a while during migration.

  ✓ Verified by official sourceSource: Claude Platform API sürüm notları

+6 →

New in this edition · 6

• 29 Oct
  Pricing / PlanChatGPT Work ve Codex

  Work and Codex usage doubled on Pro and Business Premium plans

  Work and Codex usage allowances on Pro 100 and Pro 500 plans have been doubled from their previous levels. The previously announced Pro 200 reduction was also canceled, and subscribers with lower allowances were raised to the same level.

  Usage details

  Who can use it: ChatGPT Pro 100, Pro 200, and Pro 500, and Business Premium; depends on the plan and account status.

  How to use it: No action is required; the new usage allowance is applied to your account automatically.

  Tip: Instead of saving intensive Work or Codex tasks for the end of the month, use the new allowance for longer-running tasks.

  ✓ Verified by official sourceSource: ChatGPT yayın notlarıSource: ChatGPT Pro kademeleri
• 29 Oct
  ModelChatGPT

  GPT-6 Luna gives shorter, less sycophantic responses

  OpenAI updated GPT-6 Luna in ChatGPT to provide shorter responses and reduce overly affirming language. The update aims for more direct and balanced results in everyday conversations.

  Usage details

  Who can use it: In the ChatGPT Chat experience, depending on account and phased rollout status.

  How to use it: In ChatGPT, use the GPT-6-based Chat experience from the model selector; no additional setting needs to be enabled.

  Tip: To make responses even shorter, specify the target length and format clearly in your prompt.

  ✓ Verified by official sourceModels: GPT-6 LunaSource: GPT-6 Luna 29 Ekim güncellemesi güvenlik ek notu
• 28 Oct
  AgentDots

  Texting with Dot via iMessage or RCS begins

  The feature, which is starting as a beta for Pro users in the US, lets users reach dot via iMessage or RCS. Dot can also send users proactive updates; memory and plugin context are shared with dot in ChatGPT.

  Usage details

  Who can use it: Phased beta rollout to Pro users in the US.

  How to use it: Open the dot profile in ChatGPT and select Texting; enable iMessage on iPhone or RCS on Android.

  Tip: Review your dot instructions and connected plugins to limit proactive updates sent by message.

  ✓ Verified by official sourceSource: ChatGPT yayın notları
• 28 Oct
  Pricing / PlanOpenAI API

  Monthly promotional API credits for eligible Pro and Business subscribers

  Eligible ChatGPT Pro and Business subscribers can activate monthly promotional API credits to develop apps and agents. The credits are separate from ChatGPT usage and are designed to be used in the connected API organization.

  Usage details

  Who can use it: Eligible ChatGPT Pro and Business subscribers; credit amount and eligibility vary by account and offer.

  How to use it: Connect your own API organization in ChatGPT or the API Platform and activate the subscription benefit.

  Tip: Use the credits first for low-risk prototypes, and check the API organization's billing settings separately.

  ✓ Verified by official sourceSource: OpenAI API changelogSource: ChatGPT yayın notları

+2 →

New in this edition · 2

• 27 Oct
  AvailabilityMuse

  Muse launches in Singapore

  Meta has launched its personal agent, Muse, in Singapore. Users can direct shopping, dining, travel and everyday tasks through connectors for Gmail, Google Calendar and Instagram, as well as Shopee, Grab, Singapore Airlines, Agoda, ShopBack, Klook, Trip.com and Lazada.

  Usage details

  Who can use it: Available in Singapore on iOS, Android and muse.ai; basic use is free, with subscription plans available for higher usage.

  How to use it: Open the Muse app or muse.ai and connect the relevant apps and connectors in settings.

  Tip: Before making a purchase or reservation, ask Muse to compare prices, options and terms.

  ✓ Verified by official sourceModels: Muse SparkSource: Muse Is Now Available in Singapore: The World’s First Personal AI Agent Built for Everyone
• 26 Oct
  Plugin / IntegrationSocial Technologies MCP

  Create and configure a Meta app with an agent

  Social Technologies MCP has moved beyond providing API and integration information and can now handle key steps in the app lifecycle through an agent. The agent guides users through creating an app, finding or creating a Business Portfolio, linking it, and adding use cases such as WhatsApp Business messaging.

  Usage details

  Who can use it: Rolling out gradually; it may not be available to all developers at the same time.

  How to use it: Reconnect the MCP connection and ask the agent you use to create a new Meta integration; approve write operations one by one.

  Tip: Do not share your app secret in the agent chat; Meta requires you to retrieve it from App Dashboard.

  ✓ Verified by official sourceSource: Developers can now create and configure an app for Social Technologies Dev Center using their agent

New in this edition · 15

• 30 Oct
  RetiredKimi API

  New registration and retirement schedule announced for the K2 series

  The official pricing document states that, starting October 31, 2026, K2 series models will no longer be available to newly registered users, and that all K2 models are planned to be retired by the end of December 2026.

  Usage details

  Who can use it: Existing users are unaffected during the transition period; K2 access will be closed to newly registered users after October 31.

  How to use it: Migrate API applications using K2 to the Kimi K3 or Kimi K3.1 Pro model IDs.

  Tip: Test the migration before the end of December by changing the model ID in the configuration file.

  ✓ Verified by official sourceModels: Kimi K2.6, Kimi K2.7 CodeSource: Model Inference Pricing
• 30 Oct
  SettingKimi Code CLI

  Single-file read limit reduced to 100,000 characters

  In Kimi Code CLI 2.4.1, the amount of a file the tool can read at once is capped at 100,000 characters, or approximately 25,000 tokens. Users who want a higher limit can increase it again through configuration.

  Usage details

  Who can use it: Kimi Code CLI 2.4.1 users.

  How to use it: Increase the max_chars value in the [read] section of config.toml.

  Tip: For large log or data files, increase max_chars carefully to preserve context instead of splitting the file into chunks.

  ✓ Verified by official sourceSource: Kimi Code CLI Changelog
• 28 Oct
  Plugin / IntegrationKimi Code CLI

  Importing Claude Code and Codex sessions added

  Kimi Code CLI 2.4.0 added a feature that converts Claude Code and Codex sessions from the past 15 or 30 days into regular Kimi Code sessions, letting you pick up where you left off.

  Usage details

  Who can use it: Kimi Code CLI 2.4.0 users.

  How to use it: Run the /import-sessions command or the kimi import-sessions command, then select the time range.

  Tip: Back up the old tool’s session folders before migrating; import only the date range you need.

  ✓ Verified by official sourceSource: Kimi Code CLI Changelog
• 28 Oct
  SettingKimi Code CLI

  Plan mode now starts only when enabled by the user

  With CLI 2.4.0, Agent can no longer switch to Plan mode on its own; every plan awaits user review, and the Never Ask setting does not automatically approve plans.

  Usage details

  Who can use it: Kimi Code CLI 2.4.0 users.

  How to use it: Enable Plan mode yourself using the plan setting in the session or the relevant command, then approve the suggestion.

  Tip: If you want to review a plan before files are touched, this change provides a safer review workflow.

  ✓ Verified by official sourceSource: Kimi Code CLI Changelog

+11 →

New in this edition · 3

• 27 Oct
  APIGrok API

  Custom moderation guidelines in API requests

  The `guidelines` array added to Grok API requests lets you define `decline` rules that make the model stricter in the nsfw, violence, and profanity categories. This allows developers to set application-specific safety boundaries in addition to default moderation.

  Usage details

  Who can use it: Available to API teams using Chat Completions and Responses API; unavailable for multi-agent models and some deferred requests.

  How to use it: Add the `guidelines` array to the request body; for each item, send the `type: "policy"`, relevant `category`, and `behavior: "decline"` fields.

  Tip: Send only the categories you want to change; unspecified categories retain their default behavior.

  ✓ Verified by official sourceModels: grok-4.7Source: Custom Moderation Controls
• 24 Oct
  SettingGrok Bot

  Auto-review rules can be managed from cards

  Grok Bot can now suggest adding, changing, or removing an Auto-review rule from the in-app approval card. Because Auto-review performs a second check, it also reduces unnecessary approval cards for actions deemed safe.

  Usage details

  Who can use it: For accounts using Grok Bot; the relevant desktop app version must be 0.75.2 or later.

  How to use it: Open and approve the Auto-review rule suggestion shown on an approval card; rules can also be edited in Grok Bot settings.

  Tip: Add rules for repeated, low-risk actions, but keep irreversible actions such as sending money or external messages subject to manual approval.

  ✓ Verified by official sourceSource: Grok Bot Changelog
• 23 Oct
  FeatureGrok Bot

  Vault, email forwarding, and higher routine limits

  Bots can now save login details, cards, and contact information to Vault in Settings and use them to fill out forms; they can forward emails from their own inboxes, preserving formatting, images, and attachments. The total routine run limit has increased from 100 to 300 per hour, and the per-routine limit is now 200.

  Usage details

  Who can use it: For Grok Bot users; Vault and bot email features are available in version 0.75.2 or later.

  How to use it: Open Vault from Settings; ask a Bot to fill out a login or payment form, or forward an email from its own inbox.

  Tip: Save only the information in Vault that automation actually needs, and keep Auto-review approval enabled for payment submissions.

  ✓ Verified by official sourceSource: Grok Bot Changelog

New in this edition · 1

• 27 Oct
  ModelQwen-Audio

  Qwen-Audio-3.1-Realtime-Flash released

  Qwen-Audio-3.1-Realtime-Flash offers bidirectional, real-time conversation with text and audio input and output. Support for function calling, web search, and voice cloning, along with a 262.144-token context window, delivers a significant upgrade for voice agents.

  Usage details

  Who can use it: Users covered by Model Studio’s International scope; API usage is subject to charges.

  How to use it: Select qwen-audio-3.1-realtime-flash as the model ID in Model Studio and connect to the real-time audio API.

  Tip: Enable function calling for voice customer service and have the agent confirm the outcome of an action during the conversation.

  ✓ Verified by official sourceModels: qwen-audio-3.1-realtime-flashSource: Model lifecycle and updates

No notable change since the last edition.

Latest updates

• 20 Oct
  Plugin / IntegrationDeepSeek Recipe

  DeepSeek Recipe 0.2.0 released

  DeepSeek Recipe 0.2.0 updated its Rust libraries and Python bindings to convert V4 and V4.1 requests based on the model version. The new release handles image, tool search, structured output, and reasoning options more accurately across Responses API, Chat Completions, and Messages streams.

  Usage details

  Who can use it: Available to developers as an open-source Rust/Python package.

  How to use it: Upgrade to deepseek-recipe 0.2.0; use ConversionOptions.model_versions_by_name or with_model_alias for custom model names.

  Tip: When migrating from 0.1, check the changed APIs for ConversationRequest and stream generators.

  ✓ Verified by official sourceModels: DeepSeek-V4-Pro, DeepSeek-V4.1-FlashSource: DeepSeek Recipe değişiklik günlüğü
• 17 Oct
  AgentDeepSeek app/chat

  The 1017 release focused on end-to-end project generation

  DeepSeek-V4.1-Flash-1017 strengthened end-to-end project generation for tasks such as game development and 3D modeling through self-feedback reinforcement learning. DeepSeek showcased a web-based racing game, a Portal 2 remake built with Godot, and a 3D model of the Forbidden City.

  Usage details

  Who can use it: DeepSeek web/app and compatible agent environments using the model.

  How to use it: Specify the complete project goal, tools to use, and delivery criteria in a single prompt, and run it in an agent environment with file access.

  Tip: Clearly define the working directory, dependency rules, and test commands before generating code and assets.

  ✓ Verified by official sourceModels: DeepSeek-V4.1-Flash-1017Source: DeepSeek-V4.1-Flash-1017 duyurusuSource: DeepSeek güncellemesini aktaran Silicon Republic haberi
• 17 Oct
  ModelDeepSeek open weights

  DeepSeek-V4.1-Flash-1017 open weights released

  DeepSeek released the V4.1-Flash-1017 weights on Hugging Face and ModelScope. The Hugging Face model card lists the release as MIT-licensed and usable with image-to-text and text generation workflows.

  Usage details

  Who can use it: Developers with access to Hugging Face or ModelScope; local execution requires suitable hardware and inference infrastructure.

  How to use it: Load the deepseek-ai/DeepSeek-V4.1-Flash-1017 model from Hugging Face with Transformers, vLLM, or SGLang.

  Tip: Factor the model card’s approximately 763B parameters and 1M context requirements into your hardware plan.

  ✓ Verified by official sourceModels: DeepSeek-V4.1-Flash-1017Source: DeepSeek-V4.1-Flash-1017 duyurusuSource: DeepSeek-V4.1-Flash-1017 model kartı

No notable change since the last edition.