Skip to content

v1.57.0

Choose a tag to compare

@github-actions github-actions released this 10 Nov 02:39
· 302 commits to main since this release

Model per message picks the effort too, routes subagents, and thinks harder while tool calls fail

  • Effort per message. Each message now gets a model and an effort, from a range per model:

    • Haiku: low or medium
    • Sonnet: low to xhigh
    • Opus: medium to max

    The local proxy writes the effort into the request. The prompt cache is kept per model and effort, so a change of effort is weighed like a change of model: it is made only when it still pays. A model's own default is never written.

  • Subagents are routed on their own. A subagent's first request is held for up to 300 ms while its task is read, then sent on the model and effort that task needs (a file search on Haiku, a design review on Opus). Its later requests stay on that choice.

  • More thought while tools fail. Two failing tool turns in a row raise the effort one level, and four raise it two. The effort never goes above xhigh, and this never applies to a conversation over about 600,000 characters. The raise lasts until your next message, and the pane shows "more thought". The model is never changed for this.

  • Refusals: a refused effort is retried as Claude Code asked for it, and that change is not tried again in that terminal.

CI

  • The full e2e suite is pinned to ubuntu-24.04 (ubuntu-latest is now 26.04, which has no Playwright Chromium).
  • The Linux release build is pinned to ubuntu-24.04, so the minimum glibc version stays where it was.

Full Changelog: v1.56.2...v1.57.0