Highlights: Reliable ChatGPT Chat/Work automation and accurate provider reasoning-token accounting.
- Browser: support ChatGPT's Chat/Work composer, model and effort controls, turn identity, Markdown and file capture; preserve Work-conversation guards, paste multiline prompts intact, wait for attachment hydration without duplicate uploads, and persist new macOS manual-login profiles with the native Keychain; fixes #517, thanks @Cjschmi2, @moeuu, @hongho55, @StartupBros, @lifeofgurpreet, @rugnasyab, @cafeSowoo, and @Gerry9000.
- Browser: warn when a configured Chrome executable cannot replace a running shared profile, explain how to switch safely, and share reuse handling with Project Sources; fixes #510, thanks @Sogl.
- API: include Gemini thinking tokens in billed output, cost, and token totals, bill xAI reasoning tokens as output, and show reported reasoning tokens for OpenAI Responses and custom-gateway runs; thanks @devYRPauli.
- Dependencies: update pinned Hono to 4.13.11, its Node adapter to 2.1.3, and pnpm to 11.28.2 while retaining the two-day release-age policy and Node >=24.
- Dependencies: update MCP server/client to 2.2.0, the legacy MCP SDK to 1.31.0, oxfmt to 0.71.0, and oxlint to 1.86.0 within the two-day release-age policy; retain Node >=24.
- Dependencies: refresh provider/MCP SDKs, cookie and browser tooling, test tooling, and pnpm within the two-day release-age policy; update dotenv's compatible preload entrypoint, retain Node >=24, and keep Undici on v7 for native-fetch compatibility; thanks @dependabot.