llm-anthropic 0.25 adds Claude Opus 4.7 support with extended thinking capabilities

Simon Willison released version 0.25 of llm-anthropic, a plugin for the open-source llm command-line tool that provides access to Anthropic's Claude model series. The release adds support for Claude Opus 4.7, a new model variant that accepts a higher thinking_effort setting of xhigh for extended reasoning tasks.

New boolean configuration options shipped with this release allow users to control how extended thinking outputs are displayed. The thinking_display option enables summarized reasoning output in JSON formats and logs, while thinking_adaptive provides adaptive behavior during inference. Willison's notes indicate that thinking_display output currently works only with JSON output modes.

The release raised default token limits across all supported models to match their maximum thresholds, reducing friction for users who previously needed manual configuration to reach optimal output lengths. Additionally, the update removes references to an obsolete beta header for structured outputs that was used with older Claude model versions, reflecting changes in Anthropic's API maturity.