AI Model Tracker
All tools
Anomaly

OpenCode CLI

2 days since its last release · 2.2.7 ·

OpenCode is an AI coding tool from Anomaly. Its latest release is 2.2.7, released on 10 Nov 2026. It shipped 18 releases in the last 30 days, and we have tracked 984 since 12 Jun 2025.

Latest
2.2.7
10 Nov 2026
Last 30 days
18
releases
Usual gap
~1d
between releases
Tracked
984
since 12 Jun 2025
Homepage
opencode.ai

Releases per month

last 12 months
  1. 96 Dec
  2. 50 Jan
  3. 32 Feb
  4. 25 Mar
  5. 34 Apr
  6. 35 May
  7. 16 Jun
  8. 19 Jul
  9. 15 Aug
  10. 27 Sep
  11. 22 Oct
  12. 2 Nov

Release history

984 releases

Releases 48

  1. 2.2.7 Notable

    GitHub Copilot Auto model routing and new tab shortcuts

    • Pick Auto in the model picker to let Copilot's router choose the model per session at the Auto discount, with fallback to the default chat model
    • Close tabs to the right, new close and reopen tab shortcuts, and Mark as read or unread on tabs, Home rows, project rows and inbox cards
    • Remote MCP logins with rotating refresh tokens no longer sign you out during concurrent refreshes or brief token endpoint failures
    • read accepts a negative offset to read from the end of a file and suggests close names when a file is missing
    • Breaking: ImageResponse.image is removed in @opencode/ai, and several SDK endpoints no longer wait for plugins to load
    Release notes
  2. 2.2.5 Notable

    Keyboard shortcuts for worktrees, base branches and shell mode

    • Desktop adds shortcuts to choose a worktree (Cmd+Opt+O or Ctrl+Alt+O) and its base branch, and Cmd+Shift+X or Ctrl+Shift+X toggles shell mode
    • Hold Cmd or Ctrl in the quote popup, or press Cmd+Enter or Ctrl+Enter, to send just that quote and its comment
    • The compaction buffer in opencode.json can now be set as a percentage of the model's context window, such as 20%
    • Clearer read and edit errors name the missing folder or show the file's closest matching lines
    • MCP servers sending stdio messages over 10 MiB no longer disconnect, and LiteLLM context window errors now trigger compaction
    Release notes
  3. 2.2.4 Notable

    Desktop terminals now persist across restarts, synced with the TUI

    • Desktop terminals belong to the session, survive server restarts with scrollback and names, and stay synced with the TUI and other windows
    • Experimental artifacts let agents publish, update and serve HTML or static artifacts via a built-in local provider or plugin storage
    • Guided Claude Agent setup walks through installing Claude Code, signing in and verifying the connection
    • New Choose base branch shortcut when creating a worktree, also in the command palette
    • Enabled schedules now run under opencode serve and fire only once across multiple server processes
    Release notes
  4. 2.2.3 Notable

    Use your Claude Code login and models through a new Claude Agent provider

    • New built-in Claude Agent provider: install Claude Code, run claude auth login, then pick Claude Agent in /connect
    • Desktop does much less rendering work for streaming replies and long sessions
    • Desktop fixes for side-panel tabs, repeated Session not found toasts and Open file over HTTP
    • Breaking: SDK media APIs moved to @opencode/ai/experimental, TUI plugins need OpenTUI 0.8, and integration.use no longer covers MCP, skills or plugins
    • Desktop adds a spellcheck toggle, PageUp and PageDown timeline paging, and separate notifications for questions
    Release notes
  5. 2.2.2 Notable

    Image generation now on by default when a provider supports it

    • Agents can use image generation with no extra setup when a connected provider supports image models, with the preferred model set via config
    • Clicking the Git branch on the new-session screen now opens the project's worktree picker directly
    • Timeline quotes, terminal quotes and browser annotations now share one compact optional-comment editor
    • Desktop restores classic question cards by default, with the redesigned question UI available as an experimental setting
    • SDK schedules gain if and until inputs, and Go and ChatGPT usage-limit messages now show reset times
    Release notes
  6. 2.2.1 Notable

    New /goal command keeps sessions working until an objective is met

    • /goal <objective> keeps a session running until a reviewer confirms the goal is met, with pause, resume and clear controls in the TUI and Desktop
    • Opt-in imagegenerate tool lets agents create and edit images, with GPT Image 2, MiniMax and Meta Muse support
    • Opt-in scheduled prompts via /schedule, agent tools and new SDK schedule methods
    • Desktop makes the redesigned searchable model picker the default and lets you view live output from background shells and stop them
    • Breaking for @opencode/ai imports: fal, Black Forest Labs, Replicate and Stability providers and five legacy entry points removed
    Release notes
  7. 2.2.0 Notable

    Experimental desktop plugins arrive alongside a refreshed desktop app

    • Create, install and share desktop plugins (experimental), managed from Settings › Extensions › Plugins
    • Refreshed desktop app with Focus and Pro layouts, rewind or fork from any message, and a searchable model picker
    • Experimental image and decision generation APIs for plugins and the SDK
    • Breaking: @opencode/ai root now exports the Promise API, the GitHub Action is removed, and TUI plugins need OpenTUI 0.7
    • Adds configurable prompt cache TTLs per provider and model, plus GPT-6.1 Terra and Terra Fast via Sign in with ChatGPT
    Release notes
  8. 2.1.4 Notable

    MCP input forms now appear instead of timing out, plus background compaction

    • MCP servers that ask for input during a tool call now show a form instead of waiting until timeout, including in desktop and web sessions
    • Experimental background compaction prepares a summary early so automatic compaction does not wait, though unused summaries still cost tokens
    • When a compaction request goes over the provider limit, older tool output is shortened first so more useful history is kept
    • New experimental TUI tools option lets agents open and type into the session terminal without taking keyboard focus
    • New heap capture tools: opencode debug heap-dump for running servers and Capture heap snapshots in the Desktop command palette
    Release notes
  9. 2.1.3 Notable

    Bun 1.4.4 security fixes, faster Desktop highlighting and voice SDK

    • Bundled Bun runtime upgraded to 1.4.4 with security fixes for Brotli, TLS, HTTP, WebSocket and fetch, plus updated root certificates
    • Desktop code blocks, file views and review diffs now use Pierre Highlights for faster highlighting and lower memory use on large files
    • SDK adds client.voice.models() to list experimental voice models and supported WebRTC or WebSocket transports
    • Voice sessions can use GPT-Live hosted by a connected OpenCode Console workspace, with instruction updates via client.voice.instructions()
    • Fixes for custom provider baseURL overrides, incomplete compaction summaries, TUI crashes when file watchers are unavailable and MCP media handling
    Release notes
  10. 2.1.2 Notable

    Clearer free-limit handling with reset times and upgrade options

    • When free messages run out, Desktop and the TUI now show the reset time, restore your prompt and offer upgrade or model-switching actions
    • New OPENCODEDISABLEFREEMODELS setting hides zero-cost models from catalogs, defaults and fallbacks
    • New experimental.locationidletimeout setting controls when inactive location services unload, or turns unloading off
    • Installer now creates a missing zsh or Fish config file so fresh installs add OpenCode to PATH
    • SDK session errors now include retryAfterMs so clients can show provider retry timing
    Release notes
  11. 2.1.0 Notable

    Persistent Windows terminals, experimental todo lists and SDK breaking changes

    • Windows x64 and ARM64 now get persistent terminals, matching the workflow on macOS and Linux
    • Experimental todo lists let models track tasks in the session and sidebar when enabled in config
    • Breaking SDK, event and plugin changes rename provider replay data to native and move credential setup answers to metadata
    • Long-running TUI commands and shell calls no longer drop past the five-minute fetch deadline
    • Session archiving is back, saved permissions can be edited, and summary compaction now covers the whole conversation
    Release notes
  12. 2.0.32 Notable

    /copy now grabs only the latest reply, plus headless login fix

    • /copy in the TUI now copies the latest assistant message or compaction summary; the full transcript is still available via the command palette or /export
    • opencode auth login no longer crashes in containers or headless environments without a browser; use the printed URL to continue
    • Experimental SDK voice session API for OpenAI Realtime and GPT-Live sessions and xAI Grok Voice sessions, using server-stored API keys
    • Gemini sessions now retry or continue after intermittent MALFORMEDRESPONSE finishes instead of failing
    • Snapshot capture on Windows handles concurrent Git object writes more reliably
    Release notes
  13. 2.0.30 Notable

    Per-turn changed files, docked TUI diffs and keyboard tab switching

    • Turn on Settings → Preferences → Timeline → Changed files to see each completed turn's changed files, line counts and expandable diffs
    • Running /diff in the TUI now opens the diff viewer beside the session, press f to switch to full screen, and the layout fits narrower terminals
    • Move between session tabs with Option+↑/↓, or add Shift to jump between unread tabs
    • Removed the retired session-sharing backend and the /share command
    • Instructions now load local glob patterns and remote URLs, plus fixes for MCP compatibility, compaction limits and TinyFish web search
    Release notes
  14. 2.0.29 Notable

    Remote device pairing and multiple provider accounts

    • Pair a device from Settings → Pairing to turn on OpenTunnel remote access, with a QR code or a link you can copy
    • Switch, rename, remove or add provider accounts from the connect dialog or Settings → Providers
    • Responses that hit their output-token limit now continue automatically instead of stopping partway
    • The desktop session view now shows context tokens and total cost, and image previews support pinch-to-zoom and panning
    • Access policies move to a top-level policies config key, and plugins can now use ctx.permission.assert
    Release notes
  15. 2.0.28 Notable

    Per-project auto-accept, native Windows ARM64 and WinGet installs

    • Per-project auto-accept permission overrides in Settings → Projects, toggled with mod+shift+a in a session
    • Native Windows ARM64 builds for downloads, CLI installs and desktop updates instead of x64 emulation
    • Official WinGet packages for the CLI and desktop app, with upgrades and uninstalls handled through WinGet
    • Long Gemini and Vertex Gemini responses split by the provider now continue automatically as one response
    • TUI shows release highlights once after an update, reopenable with /highlights; optional official Figma plugin added
    Release notes
  16. 2.0.27 Notable

    Built-in worktree tools, safer remote pairing and installable web app

    • New Code Mode tools let OpenCode create, list and remove isolated git worktrees, with edit approval required for removal
    • The first opencode pair --remote now warns that it exposes your server and asks for confirmation; use --yes for scripts
    • The web app can be installed from Chrome or Edge, including behind authenticating proxies
    • Desktop and TUI timelines now show execution errors that happen before or between model steps
    • Projects whose directories no longer exist are hidden automatically, and the ChatGPT OAuth input limit for gpt-6.1-sol rises to 500,000 tokens
    Release notes
  17. 2.0.26 Notable

    Vertex sign-in with Google Cloud credentials and faster snapshots

    • Vertex can now connect with Google Cloud credentials (gcloud auth or environment), letting you choose the project and location
    • Session snapshots add less overhead, recover from interrupted or concurrent captures, restore faster and use less disk
    • Transcript links now open when clicked, including wrapped URLs, and hyperlinks work in more terminals, with FORCEHYPERLINK=1 to force them on
    • Step 5 Preview Free added to the free models for a limited time, replacing the retired Fledge Alpha Free
    • Remote-access URLs from opencode service now use a persistent random subdomain instead of a predictable route
    Release notes
  18. 1.18.30 Notable

    New Astra system prompt for GPT-6 models, plus provider fixes

    • Added the Astra system prompt for GPT-6 models
    • Added reasoning effort variants for supported GitLab GPT and Claude models
    • Bedrock DeepSeek model IDs, including ARN-based IDs, now resolve correctly
    • Updated the Azure and OpenAI provider SDKs to pick up compatibility fixes
    Release notes
  19. 1.18.24 Notable

    Azure providers can now sign in with Entra ID via the Azure CLI

    • Azure providers can sign in with Microsoft Entra ID through the Azure CLI instead of needing an API key
    • V1 now reads supported V2 config fields, so newer config files keep working in more mixed setups
    • Bedrock reasoning responses no longer get cached as empty messages that can't be replayed
    • Desktop: archived sessions now disappear from the Home list immediately
    Release notes
  20. 1.18.19 Notable

    Cloudflare AI Gateway passthroughs and Codex limits that match ChatGPT plans

    • Added native OpenAI and Anthropic passthroughs for Cloudflare AI Gateway models
    • Codex rate limits now match ChatGPT subscription limits more closely
    • Web search now works with the OpenCode Go provider
    • Removed built-in Qwen sampling defaults that could send unsupported settings
    • Fixed authenticated providers not showing in /connect, and malformed model pricing no longer breaks usage cost calculation
    Release notes
  21. 1.18.15 Notable

    Desktop app gains JSON transcript export and broader language support

    • Export full session transcripts as JSON from the desktop app
    • Much broader locale coverage in the desktop app, with missing translations filled in
    • Message ordering, revert and fork now follow real chronology, even with out-of-order or imported message IDs
    • Repeated compaction now keeps earlier tool-call history in summaries
    • Blob-based attachments now load correctly in the web UI
    Release notes
  22. 1.18.14 Notable

    Simpler xAI login and more automatic retries on provider errors

    • xAI login is now a single device-code flow that works better in headless and remote environments
    • More transient provider and network errors are retried instead of failing immediately
    • Structured mid-stream provider errors are kept so compatible providers can retry failed responses
    • Remote workspaces now resolve prompts from the remote project root and log upstream 5xx response bodies
    • ACP usage totals now count cache writes, and turns wait for queued session updates before ending
    Release notes
  23. 1.18.13 Notable

    Desktop app gets early right-to-left layout support and more languages

    • Early right-to-left layout support added across the desktop app and session views, with fixes to tabs, drawers, resizing and the titlebar
    • More desktop UI copy localized, including native menus, update dialogs and help content, plus more supported languages
    • Locale-aware plural rules so translated counts and summaries read correctly
    • Markdown parsing moved off the main thread to keep the desktop app responsive while rendering messages
    • GitHub pull request reviews in the TUI now include the pull request number and URL in context
    Release notes
  24. 1.18.10 Notable

    Automatic Modal model discovery and desktop app polish

    • Available Modal models are now discovered automatically
    • Desktop toast notifications stack, dismiss and fit mobile screens better
    • The same attachment can no longer be added more than once, and the new session button is always visible
    • Broken saved tab state no longer carries over between desktop launches
    • Model variant selector stays hidden until model data has loaded
    Release notes
  25. 1.18.9 Notable

    Opt-in desktop sidecar, collapsible provider settings and fixes

    • Added an opt-in V2 desktop sidecar backed by the bundled CLI service
    • Model provider sections in V2 settings can now be collapsed
    • Restored compatibility with legacy MCP SDK clients
    • Fixed a desktop crash that could break navigation and improved home session list loading
    • Removed extra vertical borders from the V2 home projects view
    Release notes
  26. 1.18.8 Notable

    Better MCP server and OAuth compatibility, plus desktop fixes

    • Improved compatibility with newer MCP servers and OAuth flows
    • MCP servers reconnect after expired SDK sessions, including concurrent requests
    • Desktop adds a shortcut to open the project selector in new sessions
    • Fixes desktop freezes when closing projects and while the right panel loads
    • Stops sending deprecated sampling defaults to newer Gemini models
    Release notes
  27. 1.18.5 Notable

    Desktop app now works with current servers as well as legacy ones

    • Desktop app detects legacy and current servers and works with both
    • Desktop app supports current-server sessions, prompts, commands, timelines, review data, event streaming and terminal
    • Desktop app no longer auto-accepts config permissions on current servers
    • Fixes for Claude adaptive thinking, OpenAI Responses conversations, and Mistral reasoning history and prompt caching
    • Grep search results now keep symlink paths, and MiniMax M3 thinking variant selection is fixed
    Release notes
  28. 1.18.4 Notable

    Rebuilt desktop prompt input and smarter Kimi reasoning controls

    • Desktop v2 prompt input rewritten for more reliable commands, context, shell, attachments and history
    • Kimi models on Anthropic-compatible providers now use adaptive thinking with summarized reasoning by default
    • Desktop review panel keeps file tabs aligned with the diff view and resizes more smoothly
    • Embedded terminal theme now follows the app theme
    • Fixes Azure Cognitive Services endpoints, OpenAI header timeouts and provider reasoning options
    Release notes
  29. 1.18.3 Notable

    Desktop command palette can now search and open sessions

    • Home page command palette search in the desktop app can now find and open sessions
    • Up Arrow now closes the subagent picker when the first item is selected
    • Fixed home page scrolling so sticky headers and the session list behave correctly
    • Desktop app now waits for WSL server loading before reporting ready
    • Fixed custom agent selector visibility and removed the inactive help button
    Release notes
  30. 1.18.2 Notable

    Subagents no longer spawn nested subagents by default

    • Subagents no longer launch nested subagents by default, with a configurable subagentdepth limit when needed
    • Improved default reasoning depth for Meta models
    • Desktop adds Mod+N as another shortcut for opening a new tab
    • Desktop fixes: Help button restored in release builds, sessions with null archive times stay visible on the home list
    • Drawer close button hidden on Windows, where it clashed with the window chrome
    Release notes
  31. 1.18.0 Notable

    Desktop v2 layout completed, with onboarding and a switch back to the old layout

    • Desktop v2 migration completed, with upgrade handling for the new layout and first-launch onboarding
    • New setting to switch between the new and old Desktop layouts during the transition
    • Home loads much faster from a cold start, and the timeline loads more history at once while scrolling
    • Permission auto-accept now works for remote sessions and is kept separate for each server
    • Fixes for timeline resync after reconnecting, terminal tab focus, project picker crashes and composer caret position
    Release notes
  32. 1.17.20 Notable

    Azure AI support updated for GPT-5.6, plus a Responses Lite fix

    • Updated Azure AI support for GPT-5.6
    • Removed an obsolete Codex workaround that could interfere with OpenAI Luna Responses Lite requests
    Release notes
  33. 1.17.19 Notable

    Desktop interface redesign moves ahead, with model selection per prompt

    • Desktop composer now lets you pick a model for each prompt
    • Redesigned review panel with persistent file browsing, better file tabs and easier open-in-app actions
    • New temporary setting in the desktop app to switch between the old and new interface
    • Adds support for OpenAI pro reasoning mode and OAuth for Luna Responses Lite
    • xAI Responses no longer store responses by default, plus many desktop UI fixes
    Release notes
  34. 1.17.17 Notable

    Desktop gets a free-model selector and refreshed session design

    • Desktop adds a v2 free-model selector with faster access to more providers
    • Sub-agent task rows and the revert dock are restyled to match the v2 session design
    • New dismissible tabs intro popup and a refreshed help entry point in the desktop app
    • Better handling of Meta models for reasoning variants and provider requests
    • Fixed clipped descenders in model selector labels
    Release notes
  35. 1.17.16 Notable

    Desktop gets inline file browser tabs and a composer add menu

    • Desktop adds inline file browser tabs for opening and previewing files from the side panel
    • New composer add menu for files, commands, context and shell mode that keeps draft text
    • Review panel mode and selected file are now remembered per session
    • Grok models now expose reasoning effort variants, with better xAI prompt cache routing and PDF support
    • Fixes for review pane width, side-panel glitches, large patch diffs and legacy session routing
    Release notes
  36. 1.17.14 Notable

    Code mode MCP adapter, plus desktop tab and review panel upgrades

    • Added a code mode MCP adapter for running confined orchestration scripts against connected MCP tools; the execute tool is now hidden unless code mode is on
    • Desktop app can reopen closed tabs, open tabs in the background and list recently closed projects on Home
    • Overhauled the v2 review panel in the desktop app and improved performance with large review panes
    • Improved the desktop app's integrated terminal, model search in pickers and provider connect flow
    • Fixed GitHub Copilot model routing, paginated MCP tool catalogs, and Cerebras and OpenRouter reasoning handling
    Release notes
  37. 1.17.13 Notable

    Searchable model picker and tab previews in the desktop app

    • Desktop composer gets a searchable v2 model picker and a model management flow
    • Hovering a session tab now previews its project, path, branch and server
    • WSL server setup is simpler, with clearer distro checks and install flows
    • Each desktop window now keeps and restores its own set of tabs
    • Reasoning settings now apply reliably to OpenAI-compatible reasoning models, and GitHub Copilot follow-up requests no longer fail
    Release notes
  38. 1.17.12 Notable

    Yolo mode for auto-approving permissions, plus SDK session APIs

    • TUI adds a yolo mode that auto-approves permission requests
    • SDK gains a live event stream, access to active sessions, paged session history and permission request endpoints
    • Desktop autocompletes MCP resources, adds workspace controls for new sessions and shows token and cost totals
    • Adaptive thinking enabled for Claude Sonnet 5
    • Many MCP OAuth fixes, including reconnecting after OAuth and clearer error messages
    Release notes
  39. 1.17.11 Notable

    Roll sessions back to an earlier message, including file changes

    • Session snapshots and revert controls let you roll a session back to an earlier message, file changes included
    • Desktop adds Chrome-style tab switching with mod+1 through mod+9 and draggable tabs
    • Redesigned desktop layout gets better titlebar tabs, archived sessions and a new empty home state
    • MCP OAuth URL is now always printed so manual sign-in still works
    • Desktop fixes keep prompt drafts, dialogs and attachments tied to the right project and session
    Release notes
  40. 1.17.10 Notable

    MCP resource tools, a --mini CLI mode and a diff viewer upgrade

    • Added MCP resource read and resource template listing tools, and MCP server instructions now go into the session context
    • Added a --mini CLI mode
    • TUI diff viewer can now compare against the main branch and has a configurable keybind to open it
    • Desktop keeps a separate draft prompt for each tab and adds a session progress indicator, mobile bottom navigation and collapsible server sections
    • Added support for Opencode-managed provider integrations and a V2 plugin API for Effect and Promise plugins
    Release notes
  41. 1.17.9 Notable

    Agent step limits now end runs with a final reply instead of failing

    • Configured agent step limits now force a final text response instead of failing mid-run
    • Adds high and max thinking variants for GLM-5.2 across supported providers
    • Follow-up messages are no longer wrapped in a steering reminder, keeping prompt caching effective
    • Custom headers are now passed to Copilot model requests, and Devstral detection handles different provider ID casing
    • The TUI only shows the background subagent shortcut when the server supports it
    Release notes
  42. 1.17.8 Notable

    Faster session timelines and a batch of MCP tool fixes

    • Session timelines load much faster without flicker or scroll jumps
    • OpenAI-compatible providers now accept MCP tool schemas that previously failed validation
    • Long-running MCP tools keep their timeout alive while reporting progress, and failures show the server's error text
    • Desktop adds a Home tab toggle to switch quickly between Home and your last tab
    • Cloudflare AI Gateway now receives the configured API key correctly
    Release notes
  43. 1.17.7 Notable

    MCP servers can now receive the workspace as a root, plus plugin fixes

    • MCP servers can now receive the current workspace as a client root
    • SDK clients refresh model and provider availability when integrations change, and credential calls accept location
    • Plugin client requests reuse the active server, and plugin shell environment variables now apply to PTY sessions
    • ACP shell tool calls show the command and working directory from the start
    • Desktop new session route stays scoped to its own draft server so prompts target the right workspace
    Release notes
  44. 1.17.5 Notable

    Snowflake Cortex browser sign-in and more reliable MCP connections

    • Added external browser OAuth sign-in for the Snowflake Cortex provider
    • Expired MCP sessions now recover automatically, and closed MCP connections are cleaned up instead of lingering
    • Structured MCP tool output is now shown in a readable form
    • Commands using $ARGUMENTS no longer inject file contents twice
    • The web app now has the v2 visibility settings, plus a refreshed oc-2 color theme and better project copy and move-session flows
    Release notes
  45. 1.17.4 Notable

    MCP servers can start from a working directory, plus new auth flows

    • Local MCP servers can now start from a workspace-relative directory using cwd
    • Added connector-based authentication flows and support for stored provider credentials
    • New v2 API endpoints to create and fetch sessions, list session questions and resolve the active location
    • Responses blocked by a model's content filter now show a visible error instead of failing silently
    • Fixed Gemini tool schemas with multi-type fields, and snapshots are faster on huge Git repos
    Release notes
  46. 1.17.1 Notable

    Reference usage descriptions and a new desktop session flow

    • References can now include usage descriptions for agents, appear in the new docs and be hidden from @ autocomplete
    • Desktop adds a dedicated new-session route with draft tabs that keep your prompt and project selection together
    • MCP prompt and resource requests now respect configured server timeouts, and failed server setup no longer leaves broken connections
    • Deprecated reference config entries still load under the newer references key
    • Restored macOS auto-updates and fixed handling of Anthropic fallback responses
    Release notes
  47. 1.17.0 Notable

    Faster file search, WSL support for Desktop and sturdier MCP

    • Faster file search across large projects with new fff-backed search tools
    • Desktop on Windows can now run on WSL, with WSL server management
    • Added Cohere North model support, plus Claude Fable reasoning and a MiniMax M3 thinking toggle
    • MCP fixes: cancellations stop more reliably, large catalogs paginate fully and status messages are clearer
    • Sessions can recover once from provider context-overflow errors instead of failing immediately
    Release notes
  48. 1.16.2 Notable

    Background subagents, Snowflake Cortex provider and safer edits

    • Running subagents can now be sent to the background so you can keep working
    • Added Snowflake Cortex as a provider
    • Edit operations now refuse loose matches that could overwrite the wrong code or replace a file by mistake
    • Diff viewer adds next and previous hunk navigation, and desktop improves multi-server support with per-server session tabs
    • Fixes Bedrock sessions hanging and GPT-5 request failures from unsupported reasoning summaries
    Release notes

Smaller updates 52

  1. 2.2.6

    No release notes published.

  2. 2.1.1

    No release notes published.

  3. 2.0.31

    No release notes published.

  4. 2.0.25

    No release notes published.

  5. 2.0.24

    No release notes published.

  6. 2.0.23

    No release notes published.

  7. 2.0.22

    No release notes published.

  8. 2.0.21

    No release notes published.

  9. 2.0.20

    No release notes published.

  10. 2.0.19

    No release notes published.

  11. 2.0.18

    No release notes published.

  12. 2.0.17

    No release notes published.

  13. 2.0.16

    No release notes published.

  14. 2.0.15

    No release notes published.

  15. 2.0.14

    No release notes published.

  16. 2.0.13

    No release notes published.

  17. 2.0.12

    No release notes published.

  18. 2.0.11

    No release notes published.

  19. 2.0.10

    No release notes published.

  20. 2.0.9

    No release notes published.

  21. 2.0.8

    No release notes published.

  22. 2.0.7

    No release notes published.

  23. 2.0.6

    No release notes published.

  24. 2.0.5

    No release notes published.

  25. 2.0.4

    No release notes published.

  26. 2.0.3

    No release notes published.

  27. 2.0.2

    No release notes published.

  28. 2.0.1

    No release notes published.

  29. 2.0.0

    No release notes published.

  30. 1.18.29

    Fixes gpt-6-astra missing for OpenAI subscription users

  31. 1.18.28

    Copilot session tracking and desktop sign-in fixes

  32. 1.18.27

    Longer default timeouts for providers and Anthropic thinking fixes

  33. 1.18.26

    Fixes for Claude 5 thinking blocks and Bedrock reasoning

  34. 1.18.25

    Azure CLI sign-in no longer requires Bun

  35. 1.18.23

    Cloudflare AI Gateway fixes for third-party and Anthropic models

  36. 1.18.22

    Device login, provider compatibility and pricing message fixes

  37. 1.18.21

    Fixes for early-stopping responses and Vertex AI Gemini routing

  38. 1.18.20

    Fixes for subagent failures, network retries and provider errors

  39. 1.18.18

    Fixes for Kimi system prompt selection and xAI reasoning effort

  40. 1.18.17

    Better session compaction, capped retries and model compatibility fixes

  41. 1.18.16

    Config parsing fix and desktop app fixes for projects and macOS

  42. 1.18.12

    Fixes for Azure GPT-5.5 reasoning requests and desktop composer lag

  43. 1.18.11

    Bug fixes for MCP reconnects, reasoning model configs and desktop

  44. 1.18.7

    Desktop app bug fixes for fullscreen, command palette and menus

  45. 1.18.6

    Fixes for branch caches and provider refresh in the desktop app

  46. 1.18.1

    Desktop app fix for spacing in model provider settings

  47. 1.17.18

    Meta Muse Spark gets its own system prompt, plus a Copilot fix

  48. 1.17.15

    Desktop polish: refreshed command palette and model tooltips return

  49. 1.17.6

    Better MCP server compatibility

  50. 1.17.3

    Fixes desktop app crash introduced in the previous update

  51. 1.17.2

    Fixes for expired config logins, subagent permissions and desktop app

  52. 1.16.1

    No release notes published.

Showing the newest 100 of 984.

Choose what this site may keep in your browser. You can change it any time under Cookie settings, at the foot of every page. Privacy policy