## Core
- GitHub Copilot `gpt-5-mini` now uses low reasoning effort for better request compatibility. (@thakrarsagar)
- Workspaces now receive your auth context, so provider sign-in carries across workspace sessions.
- Cloudflare AI Gateway now drops `max_tokens` for OpenAI reasoning models so GPT-5 and o-series requests stop failing. (@kobicovaldev)
- Azure models now default `store=true`, fixing requests that require stored responses.
- Claude Opus 4.7 now supports `xhigh` adaptive reasoning. (@GrahamCampbell)
- Claude Opus 4.7 now shows summarized thinking by default.
- TUI plugins now load against the correct project when multiple directories are open.
- The bash tool uses less memory on large command output.
- Experimental workspaces now wait for sync to finish before returning writes, reducing stale reads and missed updates.
- Session restore can now replay a session into another workspace in batches.
- Sessions now retry provider 5xx errors even when the provider SDK does not mark them retryable.
## TUI
- Pasting files or large text no longer inserts content twice.
- `--agent` on the command line is no longer overwritten by the session's saved agent. (@CarloWood)
- Empty LSP, MCP, formatter, and session status responses no longer break TUI sync state.
## Desktop
- Desktop builds now show a Beta or Dev badge in the title bar when applicable.
**Thank you to 8 community contributors:**
- @ariane-emory:
- fix(core): Remove dead code and documentation related to the obsolete list tool. (#22672)
- @CarloWood:
- fix(tui): Don't overwrite the agent that was specified on the command line (#20554)
- @jerome-benoit:
- fix(nix): add shared package to bun install filters (#22665)
- @GrahamCampbell:
- feat: Add support for claude opus 4.7 xhigh adaptive reasoning effort (#22833)
- @B67687:
- fix: align stale bot message with actual 60-day threshold (#22842)
- @trbutler4:
- fix: add missing glob dependency (#22851)
- @kobicovaldev:
- fix(opencode): drop max_tokens for OpenAI reasoning models on Cloudflare AI Gateway (#22864)
- @thakrarsagar:
- fix(opencode): use low reasoning effort for GitHub Copilot gpt-5 models (#22824)