Back to rankings

Claude Code Ultrathink Now Runs a Turn at xhigh Effort on Opus 5.5, Sonnet 5.5, Haiku 5.5 and Fable 5.1

Claude CodeView original changelog

Claude Code 2.1.311 changes what the ultrathink keyword does in the interactive terminal. Typing it now runs that turn at xhigh effort on Opus 5.5, Sonnet 5.5, Haiku 5.5 and Fable 5.1 when used through Anthropic's API. If the keyword is typed while Claude is already working, it raises effort for the rest of that turn. The rainbow highlight now appears only where the keyword can actually raise effort.

Key Takeaways

  • ultrathink now means xhigh effort for the turn it is typed in, on Opus 5.5, Sonnet 5.5, Haiku 5.5 and Fable 5.1 through Anthropic's API.
  • Typing the keyword while Claude is already working raises effort for the remainder of that turn, so there is no need to wait and resend.
  • The rainbow highlight is now truthful: it appears only where the keyword can raise effort, not when the word is quoted or part of a file path.
  • This is a behavior change to an existing keyword, not a new setting, so current ultrathink habits get different results without any configuration.
  • The scope is interactive terminal sessions on Anthropic's API, so other providers and non-interactive use are not described by the release notes.
  • It ships in Claude Code 2.1.311, released October 26, 2026.

What changed

Claude Code 2.1.311 changes the behavior of ultrathink in the interactive terminal. The keyword now makes Claude run its turn at xhigh effort. This applies on Opus 5.5, Sonnet 5.5, Haiku 5.5 and Fable 5.1 when those models are reached through Anthropic's API.

The release notes list this as a change to existing behavior, so anyone who already types ultrathink into prompts will see a different result from before.

Typing it mid-turn

The keyword also works while Claude is busy. If ultrathink is typed while Claude is working, it raises effort for the rest of that turn. A user does not have to wait for the turn to finish and then resend the request with the keyword attached.

A more honest rainbow

Claude Code highlights the keyword with a rainbow effect. In 2.1.311 that highlight appears only where the keyword can actually raise effort. It no longer shows up when the word is quoted or when it is part of a file path.

This removes a source of confusion. Previously a prompt that merely mentioned the word, for example inside a path, could look as though it had switched on extra effort. The highlight now signals whether the keyword will do anything.

Who is affected

The change matters most for developers who reach for ultrathink on a hard problem and want to know what effort level they are getting. The effort level is now explicit: xhigh, on the four named models, through Anthropic's API. Sessions on other providers or models are not covered by the notes, so the behavior there should not be assumed to match.