llm-anthropic 0.26

The llm-anthropic 0.26 plugin distinguishes controlling a model's reasoning from controlling whether that reasoning is saved. In Simon Willison's release notes, the hide-reasoning option removes reasoning text from responses and logs. A separate thinking setting can disable reasoning for Anthropic's Sonnet 5 and Opus 5 models; Fable 5 always reasons.

The plugin also adopts the distinct event types introduced in LLM 0.32, allowing applications to handle reasoning, answers and tool activity separately. Hiding a reasoning trace therefore changes the returned and recorded material, without itself turning off the model's reasoning.