Skip to content
Back

AI posts

AI release notes

Official updates from Anthropic, OpenAI, Google, xAI, Mistral, DeepSeek and Meta, newest first. Open any update for a short AI summary and the full notes.

Compiled by the Toolstash team · Updated

  1. Google · New model

    New model: Gemini Robotics ER 2

    Gemini Robotics ER 2 in public preview: Released two new embodied reasoning model endpoints for robotics: gemini-robotics-er-2-preview: Advanced spatial reasoning, agentic code execution, multi-step tool orchestration, video moment finding, progress classification, and multi-robot coordination. gemini-robotics-er-2-streaming-preview: Optimized for real-time text streaming using the Live API, enabling low-latency robot agents with bidirectional audio and video input. Both model endpoints accept text, image, video, and audio inputs and support function calling with blocking behavior for physical robot actions. To get started, see the Gemini Robotics ER overview. For real-time streaming use cases, see Robotics with streaming.

  2. OpenAI · Codex app

    Sign in with ChatGPT (beta)

    Sign in with ChatGPT is beginning to roll out in beta across select plugins and partner sites, starting with Airtable, GitLab, HubSpot, Notion, Supabase, and Vercel. When you connect a supported plugin from the ChatGPT plugin directory, you can use Sign in with ChatGPT to create or link an account with that service in fewer steps. On participating partner sites, you can also select Sign in with ChatGPT to create or access your account. This makes it easier to start working with supported tools in ChatGPT and Codex. When you sign in, the partner receives only your name, email address, and profile picture, if available. You must still review and approve each plugin's requested access as a separate step.

  3. OpenAI · Codex CLI

    Codex CLI 0.146.0

    6 new6 fixes6 other

    Name new sessions with /new or /clear, pin important threads, and switch between side conversations without closing them. Support Agent Plugins manifests, workspace plugin publishing, and additional plugin marketplaces for Amazon Bedrock and Claude Code.

  4. Google · Gemini app

    Create and edit with your voice, anywhere on your desktop with Gemini on macOS

    What: We’re introducing a new way to work with Gemini using just your voice, right where you are. Press and hold the Fn key to dictate into your active window and have Gemini transcribe your spoken words into clean, polished text directly at your cursor. You can also enable Gemini reasoning in the app settings to do even more based on screen context—such as highlighting text to rewrite or reformat it, selecting local files to extract information, or generating images anywhere you want. The new feature is rolling out today in English to all users on the Gemini macOS app. Download the app at gemini.google/desktop. Why: Speaking your thoughts lets you work at the speed of thought—allowing you to ramble, hesitate, or self-correct while Gemini seamlessly makes sense of it. By taking in your screen context and outputting results directly at your cursor in any app, Gemini works right alongside you to keep you in your flow.

  5. Google · Gemini CLI

    Gemini CLI 0.53.0

    2 new4 fixes1 other

    implement LLM triage orchestrator and container build. add eval coverage report command.

  6. Google · Antigravity

    Print mode structured output formats, custom JSON schema enforcement, copyOnSelect TUI setting, and…

    Added structured output formats (json, stream-json) for print mode, support for custom JSON schema validation, enriched tool and subagent payloads, copyOnSelect configuration setting, and improved compound-command permission rules.

  7. Google · Antigravity

    Preview tabs, MCP timeouts, file attachments, and performance fixes

    Added preview tabs, attachment support for.json,.md, and.csv files, MCP timeouts, keyboard shortcuts, and various bug fixes and improvements.

  8. OpenAI · Codex app

    ChatGPT for iOS

    Voice conversations now use your selected ChatGPT voice and show usage-limit warnings. Improved task reconnection and continuity when returning to the app or unlocking with Face ID. Composer autocomplete now matches desktop plugin mentions and includes skills from installed plugins. Selected-text references remain available after sending so you can preview them again. Improved goal controls with clearer progress when pausing or resuming. Inline visualizations now render tables and visual themes more reliably. Large workspace diffs are more responsive. Fixed restored tasks changing the selected model. Prevented the composer from becoming stuck after a prompt started. Corrected browser and computer tool labels, icons, and placeholder output.

  9. Google · Antigravity

    Model-call retry/backoff configuration, LiteRT warm-up scaling, and tool execution error handling

    Release 0.1.9 of the Google Antigravity Python SDK adds model-call retry/backoff configuration, improves tool registration for custom functions, adds missing BuiltinTools exports, fixes user audio payload processing, and adds connection DebugConfig support along with ToolExecutionError handling.

  10. Anthropic · Claude Code

    Claude Code 2.1.220

    1 other

    Bug fixes and reliability improvements.

  11. Anthropic · Claude Code

    Claude Code 2.1.219

    9 new10 fixes8 other

    Added Claude Opus 5 (claude-opus-5), now the default Opus model — 1M context, fast mode at $10/$50 per Mtok. Added sandbox.network.strictAllowlist setting to deny non-allowlisted hosts for sandboxed commands without prompting.

  12. Anthropic · Claude API

    On Claude Opus 5, disabling thinking is allowed only at effort high or below: thinking: {"type": "disabled"}…

    Effort is the primary control for steering Claude Opus 5: the model supports the full ladder (low, medium, high, xhigh, max), with max for capability-critical work. Mid-conversation tool changes are now in beta on Claude Fable 5, Claude Mythos 5, Claude Opus 4.8, and Claude Opus 5: add or remove tools between turns of a conversation while preserving the prompt cache. The fallbacks parameter now supports a "default" mode, which applies Anthropic’s recommended fallback models by refusal category. We’ve removed fast mode for Claude Opus 4.7.

  13. Anthropic · Claude appsAnnouncement

    Claude Opus 5 launch

    We launched Claude Opus 5, a thoughtful and proactive model that comes close to the frontier intelligence of Claude Fable 5 at half the price.

  14. Anthropic · New modelAnnouncement

    New model: Claude Opus 5

    We’ve launched Claude Opus 5 (claude-opus-5), a step-change improvement over Claude Opus 4.8. Claude Opus 5 supports a 1M token context window (both the default and the maximum), 128k max output tokens, and thinking on by default, at $5 / $25 USD per MTok, the same pricing as Claude Opus 4.8. It’s available on the Claude API, Claude in Amazon Bedrock, Claude Platform on AWS, Claude on Google Cloud, and Claude in Microsoft Foundry.

  15. Google · Antigravity

    Markdown format custom agents, /copy index support, progressive /codesearch streaming, default temp…

    Added support for defining custom agents in Markdown format (agent.md), optional index argument for /copy, progressive streaming in /codesearch, default read access to system temporary directory, and resolved various TUI jitter, artifact viewer, and sandbox fixes.

  16. Google · Antigravity

    Permission prompts for compound commands, and fixes for disabled plugins, MCP OAuth, and clipboard corruption

    Improved permission prompts for compound shell commands, and fixed disabled plugins running hooks, MCP OAuth issuer validation, CJK clipboard copying on Windows, and /btw error on startup.

  17. OpenAI · Codex app

    ChatGPT Voice and multi-folder projects

    Powered by GPT-Live, ChatGPT Voice lets you talk through work and coordinate tasks in Chat, Work, and Codex in the ChatGPT desktop app. Start a new chat or task in voice mode, then ask ChatGPT to start, check, or steer work in other threads. On macOS, turn on Screen context to share an appshot of your frontmost window. Voice is available with Plus, Pro, Business, Edu, and Enterprise plans in the desktop app and through Remote on iOS. Local projects in the ChatGPT desktop app can now include multiple related folders. From a project's menu, select Edit project to add folders and choose the primary folder. New chats, Git operations, and automatic discovery of AGENTS.md, skills, and config.toml use the primary folder. Secondary folders remain available for file search, reading, and editing. Get started with ChatGPT Voice and multi-folder local projects. Performance improvements and bug fixes. Also includes fixes from version 26.721. Improved screen reader navigation between messages. Fixed Git detection on Windows. Fixed a freeze caused by missing icons in the Open in menu.

  18. Anthropic · Claude Code

    Claude Code 2.1.218

    4 new23 fixes10 other

    Added screen-reader announcements of deleted text for word and line deletions (Option+Delete, Ctrl+W, Cmd+Backspace, Ctrl+U, Ctrl+K) in --ax-screen-reader mode. Added HTTP status and error text to claude mcp list and /mcp when a server fails to connect, and a warning for MCP config values with hidden leading or trailing whitespace.

  19. Google · Gemini CLI

    Gemini CLI 0.52.0

    4 new6 fixes5 other

    add triage worker core foundational modules. implement octokit github action handler for egress service.

  20. Anthropic · Claude API

    You can now set an effort level on a Claude Managed Agents agent’s model configuration

    Pass effort inside the model object when you create the agent. Webhooks for Claude Managed Agents now cover the environment and memory store lifecycle: four environment.* event types and three memory_store.* event types. When creating a Claude Managed Agents session, you can now seed it with initial events. The version field is now optional when updating a Claude Managed Agents agent. Claude Managed Agents session thread event streams now support event deltas.

  21. Anthropic · Claude Code

    Claude Code 2.1.217

    3 new13 fixes4 other

    Added emoji shortcode autocomplete in the prompt input: type :heart: to insert ❤️, or :hea for suggestions — disable with the emojiCompletionEnabled setting. Added warnings when transcript writes are failing (e.g. disk full) or when session saving is off due to an inherited environment variable, instead of losing transcripts silently.

  22. OpenAI · Codex CLI

    Codex CLI 0.145.0

    6 new6 fixes4 other

    Added experimental paginated thread history with efficient resume, search, persisted names, sub-agent support, and memories. Expanded /import to migrate Cursor and Claude Code settings, MCP servers, plugins, sessions, commands, and project-scoped memories.

  23. Google · Gemini API

    Deprecated parameters: The sampling parameters temperature, top_p and top_k are now deprecated. See the…

    Deprecated parameters: The sampling parameters temperature, top_p and top_k are now deprecated.

  24. Google · Gemini app

    3.6 Flash: An upgraded model for your daily life

    What: We’re introducing Gemini 3.6 Flash, an upgraded version of our Flash model, to help you tackle everyday tasks. Try using it for quick tasks like analyzing several documents or multi-step projects like building a working prototype with the Canvas tool. We're excited to bring this model to all Gemini app users globally. To access, select "3.6 Flash" in the model drop down. Why: We want to make Gemini the most helpful AI assistant for everyone. With Gemini 3.6 Flash, you can now tackle your everyday tasks with greater confidence, without choosing between speed and quality.

  25. Google · Antigravity

    Reasoning effort command, model slug pinning, subagent model configuration, and TUI stability improvements

    Added /effort command and --effort flag for reasoning effort selection, stable model slugs, subagent model config, redesigned /model picker, scrollable /settings panel, and various MCP and background task fixes.

  26. Google · Antigravity

    Gemini 3.6 Flash default model, custom subagent instructions, and Pydantic argument coercion

    Release 0.1.8 of the Google Antigravity Python SDK updates the default text model to Gemini 3.6 Flash, adds support for custom subagent instructions, and implements automatic Pydantic argument coercion for tool calls. It also brings defensive prompt sanitization, configurable tool retries, and comprehensive stability improvements for local agent execution.

  27. Google · New model

    New models: Gemini 3.6 Flash and Gemini 3.5 Flash-Lite

    Gemini 3.6 Flash and Gemini 3.5 Flash-Lite generally available (GA): Released stable, production-ready versions of our latest 3.x Flash models: Gemini 3.6 Flash (gemini-3.6-flash): Features improved token efficiency and code/agentic planning capabilities at a lower price point than 3.5 Flash, resolving developer feedback around output verbosity. Gemini 3.5 Flash-Lite (gemini-3.5-flash-lite): Offers a low-latency, highly cost-effective subagent option designed for high-volume automation. To learn more, see the Gemini 3.6 Flash guide.

  28. Anthropic · Claude Code

    Claude Code 2.1.216

    1 new30 fixes9 other

    Added sandbox.filesystem.disabled setting to skip filesystem isolation while keeping network egress control.

  29. OpenAI · Codex app

    ChatGPT for iOS

    Added support for rendering Mermaid diagrams inline in task transcripts. Added support for interactive forms in Codex tasks. Added support for restoring unsent prompts when switching between tasks, hosts, and workspaces. Improved task lists to sort by recent activity and show unavailable hosts when creating a task. Improved the composer with selected-text previews and smoother new-task transitions. Improved goals with support for resuming blocked or usage-limited runs. Improved plan progress, Fast controls, and inline dictation. Improved Remote onboarding, composer guidance, and iPad navigation. Fixed an issue that could close the app when duplicate task-list entries appeared while starting a task. Fixed iOS 18 task actions and task-list styling. Fixed composer spacing, attachment menu padding, and duplicate transcription indicators.

  30. Anthropic · Claude Code

    Claude Code 2.1.215

    1 other

    Claude no longer runs the /verify and /code-review skills on its own; invoke them with /verify or /code-review when you want them.