Dispatch
llm-anthropic 0.26 adds Claude 5 models plus server-side WebSearch, WebFetch, CodeExecution and MCP tools
Simon Willison released llm-anthropic 0.26 on August 4, adding claude-fable-5, claude-sonnet-5 and claude-opus-5 (issues #75, #76) and exposing four Anthropic server-side tools — WebSearch, WebFetch, CodeExecution and AnthropicMCP — through LLM's `-T` flag or Python `tools=`. The plugin now requires llm>=0.32 for streaming of reasoning, tool calls and results, and collapses extended-thinking config down to `thinking` and `thinking_effort`, dropping `thinking_budget`, `thinking_display` and `thinking_adaptive`. Claude 5 models enable thinking by default; `-o thinking 0` turns it off on Sonnet 5 and Opus 5.
↳ Follow the thread