The llm-anthropic CLI tool version 0.26 introduces new models and server-side tools for web search, code execution, and the Model Context Protocol. This update adds access to claude-fable-5, claude-sonnet-5, and claude-opus-5, while replacing older command-line options with a standardised interface. The underlying engine has been upgraded to llm version 0.32, which changes how reasoning and tool results are streamed as typed events. Extended thinking parameters are now simplified into single flags, with Claude 5 models enabled by default and specific options available to disable reasoning on Sonnet 5 and Opus 5.
These changes address practical workflow issues for developers using the command line directly. The shift to server-side tools improves reliability compared to client-side implementation. Streamlined thinking parameters reduce the complexity of managing long inference chains. The new streaming format provides immediate feedback on model reasoning without requiring full response generation first.
* New models include claude-fable-5, claude-sonnet-5, and claude-opus-5
* Server-side tools cover WebSearch, WebFetch, CodeExecution, and AnthropicMCP
* Thinking parameters simplified to low, medium, high, xhigh, or max levels
