Improve Claude adaptive thinking handling across more response shapes. · Avoid OpenAI Responses phase handling that could break some conversations. · Preserve grep symlink paths in search results. (@remixz) · Preserve Mistral reasoning history across turns. · Stabilize Mistral prompt caching.
Primary source ↗Before/after verification recordings for PR #38252.
Primary source ↗Final visual evidence for PR #37967.
Primary source ↗Final visual evidence for PR #37967.
Primary source ↗Visual evidence for PR #37967.
Primary source ↗Use adaptive thinking controls for Kimi models on Anthropic-compatible providers, with summarized reasoning output by default. (@chouqin) · Reduce OpenAI provider header timeouts during slow connection setup. · Respect provider-defined reasoning options instead of falling back to the wrong reasoning controls. · Restore Azure Cognitive Services endpoint support for Azure-hosted models. · Sync the embedded terminal theme with the app theme.
Primary source ↗Added an Up Arrow shortcut to close the subagent picker when the first item is selected. · Fixed home page scrolling so sticky headers and the session list behave correctly. · Fixed startup readiness so WSL server loading is included before the desktop app reports ready. · Removed the inactive help button from the app layout. · Fixed custom agent selector visibility when custom agents are available.
Primary source ↗Stopped subagents from launching nested subagents by default, with a configurable `subagent_depth` limit when needed. · Improved default reasoning depth for Meta models. · Added `Mod+N` as another shortcut for opening a new tab. · Restored the Help button in release builds. · Kept sessions with `null` archive times visible instead of dropping them from the home list.
Primary source ↗Fixed spacing between model provider sections in Settings.
Primary source ↗Completed the Desktop v2 migration, including upgrade handling for the new layout and first-launch onboarding. · Added a setting to switch between the new and old Desktop layouts during the transition period. · Fixed file views using the wrong background in the v2 layout. · Fixed connected project picker positioning when its anchor is still loading. · Kept permission auto-accept state separate for each server.
Primary source ↗Removed an obsolete Codex workaround that could interfere with OpenAI Luna Responses Lite requests. · Updated Azure AI support for GPT-5.6.
Primary source ↗Supported OpenAI pro reasoning mode. · Disabled response storage by default for xAI Responses. (@geraint0923) · Added OAuth support for Luna Responses Lite. · Switched to another available org after logging out in the console. · Used Codex context limits for GPT-5.6 over OAuth. (@nabilfreeman)
Primary source ↗Prevent crashes and bad pricing data when GitHub Copilot returns models with a zero billing batch size. · Add a model-specific system prompt for Meta Muse Spark.
Primary source ↗Improved Meta model handling for reasoning variants and provider requests. · Fixed clipped descenders in model selector labels. · Added a dismissible tabs intro popup and refreshed the help entry point. · Updated sub-agent task rows to match the v2 session design more closely. · Restyled the revert dock in v2 sessions for clearer restore actions.
Primary source ↗Exposed reasoning effort variants for Grok models. · Improved xAI prompt cache routing and PDF file support in Responses models. · Added an Open containing folder action for projects on the home screen. · Added a composer add menu for files, commands, context, and shell mode without losing draft text. · Added a provider setup tip on new sessions when no external providers are connected.
Primary source ↗Better classify Z.ai context-window overflow errors so oversized requests surface the right failure mode (@fengjikui) · Handle unavailable config directories more gracefully when reading config files · Restore model details tooltips in the model picker (@arvsrn) · Refresh the v2 command palette to better match the rest of the new desktop UI (@arvsrn) · Make terminal shortcuts take priority when the terminal is focused
Primary source ↗Added a code mode MCP adapter for running confined orchestration scripts against connected MCP tools. · Hid the `execute` tool unless code mode is enabled. · Fixed paginated MCP tool catalogs losing tool metadata and output schema validation. · Preserved low reasoning effort for OpenRouter small-model variants instead of disabling it. · Fixed GitHub Copilot model routing to honor each model's advertised chat or responses endpoint.
Primary source ↗Force reasoning mode for OpenAI-compatible reasoning models so reasoning settings apply reliably on custom deployments · Stop replaying stale GitHub Copilot response item IDs, avoiding follow-up request failures · Let question prompts be minimized and restored without dismissing them (@arvsrn) · Clean up stale empty and old draft store files to reduce desktop state buildup · Keep pending question requests open while reviewing them, instead of dismissing them accidentally (@eXamadeus)
Primary source ↗Enable adaptive thinking for Claude Sonnet 5. · Prefer MCP content responses over structured output when both are present. · Reconnect MCP servers after OAuth even if the server was disabled. (@MaxAnderson95) · Request MCP refresh-token scope during OAuth. · Show MCP OAuth completion errors instead of a generic failure.
Primary source ↗Add session snapshots and revert controls so you can roll a session back to an earlier message, including file changes. · Always print the MCP OAuth URL so manual sign-in still works when opening the browser flow. · Add Chrome-style tab cycle shortcuts with `mod+1` through `mod+9`. · Add draggable tabs. (@arvsrn) · Improve the empty home state when you have no sessions. (@arvsrn)
Primary source ↗Added MCP server instructions to session context. (@Arcadi4) · Added Opencode-managed provider integration support. · Added MCP resource template listing. · Added MCP resource read tools. · Added a `--mini` CLI mode.
Primary source ↗Honor configured agent step limits by forcing a final text response instead of failing mid-run. · Fix Devstral model detection when provider IDs use different casing. (@Robin1987China) · Pass configured custom headers to Copilot model requests. · Add `high` and `max` thinking variants for GLM-5.2 across supported providers. (@imranshaiedi-byte) · Stop wrapping follow-up user messages in a steering reminder so prompt caching stays effective.
Primary source ↗Session timelines load much faster and avoid flicker or scroll jumps. · OpenAI-compatible providers now accept MCP tool schemas that previously failed validation. (@jquense) · Cloudflare AI Gateway now receives the configured API key correctly. (@keefetang) · MCP tools without declared schema properties now work with providers that expect object properties. · Long-running MCP tools now keep their timeout alive when they report progress. (@Nomadcxx)
Primary source ↗Plugin client requests now reuse the active server instead of assuming the default local port. · ACP shell tool calls now show the command and working directory from the start. · Plugin-provided shell environment variables now apply to PTY sessions. · MCP servers can now receive the current workspace as a client root. · MCP debug now uses the SDK's latest protocol version.
Primary source ↗Improved MCP server compatibility by declaring OpenCode's supported client capabilities.
Primary source ↗Added external browser OAuth for the Snowflake Cortex provider (@santigc6) · Improved project copy management and move-session flows in v2 · Recovered expired MCP sessions instead of leaving MCP tools disconnected · Cleared closed MCP clients so stale MCP connections do not linger · Returned structured MCP tool output in a readable form
Primary source ↗Added `cwd` support for local MCP servers so they can start from a workspace-relative directory. (@Grantmartin2002) · Added connector-based authentication flows and support for stored provider credentials. · Added v2 API endpoints to create and fetch sessions, list session questions, and resolve the active location. · File reads now return raw content with the correct content type for API and SDK clients. · MCP server log notifications are now surfaced in opencode logs.
Primary source ↗Fixed issue with 1.17.2 desktop crashing
Primary source ↗Recover from expired remote config auth by prompting you to log in again instead of failing to load config (@Ayushlm10) · Let subagents use their own configured permissions again · Restore Linux launcher and icon identity so pinned apps keep opening correctly (@neriousy) · Restore the device attachment picker (@rexdotsh) · **Thank you to 3 community contributors:**
Primary source ↗References can include usage descriptions for agents, show up in the new docs, and be hidden from `@` autocomplete when needed. · Deprecated `reference` config entries continue to load under the newer `references` config key. · MCP prompt and resource requests now respect configured server timeouts. · Anthropic fallback responses are handled correctly again. · MCP servers that fail during client setup now fail cleanly instead of leaving a broken connection behind.
Primary source ↗[github-releases] GitHub Releases
github_releases · every 180m
https://github.com/sst/opencode/releases