OpenCode v1.1.41: Model Token & Cache Fixes
OpenCode v1.1.41 fixes critical token handling bugs affecting model inference, adds provider-aware caching, and improves desktop UI responsiveness and Chinese localization.
TL;DR
- Fixed maxOutputTokens hardcoded to undefined, breaking kimi-k2.5 and other models
- Provider ID now included in SDK cache key to prevent collisions
- Copilot model variants now respect maxTokens settings
- Desktop UI improvements for mobile and Chinese localization
New
- AbortSignal support — Added to Ripgrep.files() and GlobTool for better cancellation control
- ctx.abort for grep tool — Enables graceful interruption of grep operations
- CSP media-src data: URL — Allows small audio files in Content Security Policy
Fixed
- maxOutputTokens hardcoded to undefined — Was breaking kimi-k2.5 and other inference providers; now properly respects model configuration
- Copilot model variants with maxTokens — Variants now work correctly when maxTokens is set
- SDK cache collisions — Provider ID now included in cache key to prevent cross-provider conflicts
- Desktop settings responsiveness — Mobile and small screen layouts now properly adapt
- File tab scrolling — Auto-scroll on open and mouse wheel scrolling now enabled
- Dialog alignment and padding — Fixed spacing issues in UI dialogs
- Chinese localization — Filled missing translation keys and normalized punctuation for Chinese UI
- KaTeX inline math formatting — Now properly handles punctuation following inline math expressions
- TypeScript type errors — Removed duplicate keys causing compilation failures
Update with: npm install opencode@1.1.41 or check the release page for your platform.
Source: OpenCode
