Skip to content
Agents tracked: 483 Releases (7d): 766 Agent status: no reported issuesSince last limit reset: Claude 42d · Codex 3d Updated Nov 3, 2026
AgentGid
✦ Ask

OpenCode changelog: what's new each month

Every stable OpenCode release summarised by month: the highlights, new features, improvements, fixes and anything you need to act on. 6 months covered; the current month updates daily.

October 2026

16 releases: 1.18.35 → 2.2.5

October brought major updates across OpenCode 1.x and 2.x, including experimental desktop plugins, image generation capabilities, and persistent terminal sessions. The month also marked the transition management between version lines, with 1.x staying on its release track while 2.x gained significant new features.

Action needed
  • Session sharing has been discontinued; existing share links and session-sharing commands are no longer available.
  • Automatic updates are disabled; use /update to install updates or upgrade to OpenCode 2.
  • Four provider entry points removed: ./providers/fal, ./providers/black-forest-labs, ./providers/replicate, and ./providers/stability; use nested paths instead.
  • Five legacy entry points removed with nested paths exported as alternatives.

Highlights

  • Desktop plugins are now available (experimental), allowing users to create, install, and share custom plugins.
  • Persistent desktop terminals now survive server restarts and stay synchronized across sessions.
  • Image generation is available by default when a connected provider supports it, with experimental image and decision generation APIs for plugins.
  • Claude Code integration lets users authenticate with Claude and select Claude Agent as their provider.
  • New /goal command lets sessions run toward an objective with automatic progress review until completion.

New

  • Create and share desktop plugins (experimental) with session instruction management.
  • Experimental image and decision generation APIs for plugins and SDK workflows.
  • Persistent desktop terminals belonging to sessions with scrollback surviving restarts.
  • Interactive artifacts publishing with artifact create, read, update, and delete operations.
  • Claude Agent provider with guided setup workflow.
  • /goal command to run sessions toward an objective with pause, resume, and clear controls.
  • Configurable prompt cache TTLs per provider and model.
  • Code Mode JavaScript checks including statically authorized tool calls and custom wake messages.

Improved

  • xAI tool results now include supported images while skipping unsupported formats.
  • PTY connection origin and ticket validation improved before checking existence.
  • Stats data now available in canonical redirects with JSON and Markdown formats.
  • Desktop model picker redesigned to search favorites, recent models, providers, and variants by default.
  • Long sessions in Desktop now do much less rendering work for streaming replies and conversations.
  • MCP server message handling improved to handle larger messages up to 10 MiB.
  • File edit errors now show closest matching lines or indicate the change is already applied.
  • Session media preservation when sessions are forked or imported.

Fixed

  • Failed TUI worker requests now return an error instead of hanging.
  • Upgrades for yarn-managed installs now work correctly.
  • Desktop stability issues including side-panel tabs, repeated toasts, and file-over-HTTP functionality.
  • Plugin-provided model variants are now preserved with delayed dynamic provider fallback.
  • MCP failures now preserve permission reasons and feedback, matching direct tool-call failures.
  • Persistent terminal requests with non-existent working directories are now rejected.

September 2026

9 releases: 1.18.26 → 1.18.34

September brought improved reliability across multiple AI providers, better timeout handling for slow model startups, and enhanced support for reasoning features across Claude, GPT, and other models. The month focused on fixing session persistence, provider compatibility, and request tracking.

Highlights

  • Claude 5 sessions now handle stale thinking blocks without failing when prompts or tools change.
  • Default provider timeouts increased to five minutes to reduce failures from slow model startups.
  • Session model, effort, mode, and reasoning boundaries now restore correctly when loading, resuming, or forking.
  • Improved reasoning effort support added for GitLab GPT and Claude models with proper variant handling.
  • Bedrock image attachments fixed to work only with compatible models (Claude, Nova, and Llama 4).

New

  • Reasoning effort variants now available for supported GitLab GPT and Claude models.
  • Session ID sent as GitHub Copilot interaction header to improve request tracking.
  • Desktop client ID now used during OpenCode account device authentication.
  • Gemini thinking defaults and effort options now match supported controls across model generations.
  • Namespaced session and parent-session identity headers now sent with model requests.
  • Cloudflare AI Gateway models now honor provider response and stream timeouts.
  • Support for integer GPT versions (like gpt-6) in Codex OAuth model filtering.
  • Astra system prompt added for GPT-6 models.

Improved

  • Tool call timing stays accurate when tools update their metadata while still running.
  • Bedrock reasoning and replay handling made more reliable.
  • Azure CLI sign-in now asks for resource name directly instead of querying Azure APIs.
  • Default streamed chunk timeouts set to five minutes with option to disable.
  • macOS CLI release binaries now signed with Developer ID for better compatibility.
  • Anthropic thinking block binding can now opt out via config.
  • Open-in app icon size increased for better visibility.
  • Locally compiled macOS binaries re-signed to run reliably on macOS 27+.

Fixed

  • Restored ACP session model, effort, mode, and reasoning chunk boundaries when loading, resuming, or forking sessions.
  • Session renames now save reliably from title editor and tab context menu.
  • Fixed gpt-6-astra not showing up for OpenAI subscription users.
  • Bedrock DeepSeek model IDs, including ARN-based IDs, now resolve correctly.
  • MCP browser launch failures now reported when launcher exits immediately.
  • Together AI streaming usage reporting fixed.

August 2026

15 releases: 1.18.11 → 1.18.25

August brought stability improvements across provider integrations, better support for international users, and enhanced reliability in session handling and message ordering. The month focused on fixing edge cases with various AI providers and improving the overall user experience.

Highlights

  • Fixed Azure GPT and other provider authentication issues to improve sign-in reliability.
  • Added support for exporting full session transcripts as JSON and improved session chronology handling.
  • Enhanced international language support with expanded locale coverage and right-to-left layout improvements.
  • Improved stability by fixing reconnection loops, reducing composer lag, and handling network errors more gracefully.
  • Better support for model-specific features including reasoning modes and sampling defaults across multiple providers.

New

  • Export full session transcripts as JSON from the UI.
  • Native OpenAI and Anthropic passthroughs for Cloudflare AI Gateway models.
  • Azure providers can now sign in with Microsoft Entra ID through the Azure CLI.
  • GitHub pull request reviews now include the pull request number and URL in context.
  • MERGE Gateway reasoning variants support for compatible models.
  • Early right-to-left layout support across the app and session views.
  • Locale-aware plural rules for translated counts and summaries.
  • Added locale coverage across the desktop app with updated terminology for multiple languages.

Improved

  • Moved markdown parsing off the main thread to keep the app more responsive.
  • Reduced composer lag when drafts include large pasted images or attachments.
  • Project search now matches any known recent project instead of only the first five.
  • Session compaction keeps complete recent turns and produces clearer summaries for smaller models.
  • Simplified xAI login to a single device-code flow that works better in headless and remote environments.
  • Improved retry logic with jitter to reduce repeated retry storms and handle more transient network errors.
  • Better handling of provider errors with structured mid-stream error preservation for compatible providers.
  • Sessions without titles now fall back to generated names instead of appearing blank.

Fixed

  • Fixed MCP SSE connections getting stuck in reconnect loops after server error responses.
  • Fixed provider model configs using interleaved reasoning fields with custom field names.
  • Fixed Azure GPT-5.5+ completion requests failing when reasoning is enabled.
  • Fixed several right-to-left layout issues across tabs, drawers, resizing, and titlebar interactions.
  • Fixed chronological message ordering to stay correct even when imported or legacy message IDs are out of order.
  • Fixed stale assistant errors persisting in the timeline when the same turn resumes streaming.

July 2026

19 releases: 1.17.13 → 1.18.10

July brought the completion of OpenCode's Desktop v2 migration alongside numerous improvements to model handling, reasoning capabilities, and UI reliability. The month focused on stabilizing the new interface, fixing provider-specific issues, and enhancing support for advanced reasoning models across multiple AI providers.

Highlights

  • Completed Desktop v2 migration with upgrade handling and layout toggle option during transition.
  • Improved reasoning mode support for OpenAI, Claude, Grok, Kimi, and other models with provider-specific handling.
  • Added searchable model picker with model management and free-model selector for faster provider access.
  • Fixed critical issues with MCP tool catalogs, session management, and provider routing.
  • Enhanced code mode with MCP adapter for running confined orchestration scripts against connected tools.

New

  • Code mode MCP adapter for running confined orchestration scripts against connected MCP tools.
  • Searchable v2 model picker and model management flow in the composer.
  • Composer add menu for files, commands, context, and shell mode without losing draft text.
  • Open containing folder action for projects on the home screen.
  • Inline file browser tabs for opening and previewing files from the side panel.
  • v2 free-model selector with faster access to more providers.
  • OAuth support for Luna Responses Lite.
  • Mod+N shortcut for opening a new tab and Up Arrow shortcut to close the subagent picker.

Improved

  • Force reasoning mode for OpenAI-compatible reasoning models to apply settings reliably on custom deployments.
  • Better classify Z.ai context-window overflow errors to surface the correct failure mode.
  • Improved xAI prompt cache routing and PDF file support in Responses models.
  • Exposed reasoning effort variants for Grok models.
  • Improved Meta model handling for reasoning variants and provider requests.
  • Remember review panel mode and selected file per session.
  • Rewritten v2 prompt input for more reliable command, context, shell, attachment, and history interactions.
  • Improved compatibility with newer MCP servers and OAuth flows with automatic SDK session reconnection.

Fixed

  • Stopped replaying stale GitHub Copilot response item IDs that caused follow-up request failures.
  • Fixed paginated MCP tool catalogs losing tool metadata and output schema validation.
  • Fixed GitHub Copilot model routing to honor each model's advertised chat or responses endpoint.
  • Fixed crashes and bad pricing data when GitHub Copilot returns models with zero billing batch size.
  • Fixed branch-specific repository caches so refreshing one reference no longer moves another branch checkout.
  • Fixed home session loading so the session list can update without suspending the whole page.

June 2026

15 releases: 1.16.0 → 1.17.12

June brought major improvements to session management, MCP server integration, and multi-provider support. OpenCode expanded model support across Bedrock, Snowflake, Cohere, and other providers while strengthening reliability and performance.

Highlights

  • Session snapshots and revert controls let you roll back to earlier messages including file changes.
  • Moved sessions between workspaces and directories, with managed workspace cloning that preserves your uncommitted work.
  • Faster file search across large projects with new fff-backed search tools.
  • Extended MCP server support with OAuth flows, resource templates, and improved error handling.
  • Interactive session replay with run --replay for reviewing past work.

New

  • Session snapshots and revert to roll sessions back to earlier messages with file changes.
  • Managed workspace cloning that keeps dirty and untracked files intact.
  • Moving sessions between workspaces and directories.
  • MCP resource template listing and read tools for accessing server resources.
  • Connector-based authentication flows and stored provider credentials.
  • v2 API endpoints to create and fetch sessions, list questions, and resolve active location.
  • Opencode-managed provider integration support.
  • Claude Fable reasoning support and adaptive thinking for Claude Sonnet 5.

Improved

  • File search now much faster across large projects.
  • Edit operations now refuse loose matches that could overwrite wrong code.
  • Reasoning summaries only run on providers that support them, avoiding failures on incompatible backends.
  • Session timelines load faster and avoid flicker or scroll jumps.
  • Diff viewer now supports next and previous hunk navigation.
  • Terminal themes refresh after live reloads without stale colors.
  • MCP connection status messages clearer so failures are easier to act on.
  • Desktop layout improvements including better titlebar tabs and archived sessions view.

Fixed

  • Bedrock sessions no longer hang before model responses start.
  • Anthropic fallback responses now handled correctly.
  • MCP servers failing during client setup now fail cleanly instead of leaving broken connections.
  • Gemini tool schemas with multi-type fields now work properly.
  • Content-filtered model responses now surface as visible errors instead of failing silently.
  • Session timelines no longer flicker or jump when loading.

May 2026

2 releases: 1.15.12 → 1.15.13

May brought improvements to model integrations and session management, with better WebSocket support for OpenAI responses and new metadata capabilities. Configuration handling and adaptive reasoning were also refined.

Highlights

  • ACP integrations can now send prompts, slash commands, and usage updates through acp-next
  • Sessions can store custom metadata through the API and SDK
  • Adaptive reasoning controls enabled for Anthropic Opus 4.7+ models
  • WebSocket transport added for OpenAI responses on supported channels
  • Config now loads from the opened location upward for more predictable settings

New

  • WebSocket transport for OpenAI responses (opt-in via OPENCODEEXPERIMENTALWEBSOCKETS flag)
  • Adaptive reasoning controls for Anthropic Opus 4.7+ models
  • Session custom metadata storage through API and SDK
  • Support for colons in passwords

Improved

  • Warm acp-next model and config switches now faster
  • First-session acp-next startup time improved
  • Anthropic Opus adaptive reasoning keeps summarized thinking instead of empty blocks
  • Wrapped inline tool rows stay aligned
  • Failed inline tools can expand error details in place
  • Failed OpenAI WebSocket streams retry before falling back

Fixed

  • OpenAI WebSocket response timeouts now remain active
  • Configuration loading from directory location now applies settings predictably

Summaries are written automatically from the official release notes (full changelog ↗); check the original notes before relying on a detail. OpenCode: pricing, features and alternatives · All changelogs