Agentic data science with the Data Agent Kit
cloud.google.com·7h ago
Data Science
Simon Willison's Blog·Aug 4·5 min read
llm-anthropic 0.26
TL;DR
The llm-anthropic 0.26 update introduces streaming typed events for reasoning and tool results, enhancing logging and server-side tool interactions. Key features include the ability to hide reasoning in responses and improved handling of CLI prompts.
✦ Why It Matters
Update your llm-anthropic version to leverage new features for better tool interaction and logging.
Key Takeaways
- Experiment with new models like Claude-Fable-5 for diverse applications.
- Utilize server-side tools for enhanced functionality in your projects.
- Adjust reasoning visibility to streamline user interactions.
- Simplify thinking parameters to improve model usability.
Full Summary
LLM-anthropic 0.26, released on August 4, 2026, features new models such as Claude-Fable-5, Claude-Sonnet-5, and Claude-Opus-5. This update includes server-side tools like WebSearch, WebFetch, and CodeExecution, which are accessible via the LLM's-T interface or Python tools.
The previous web_search options have been replaced with the new WebSearch tool. Additionally, the upgrade to llm>=0.32 allows reasoning, tool calls, and results to stream as typed events, enhancing the clarity of interactions.
Users can now control reasoning visibility with the --hide-reasoning flag, and the thinking parameters have been simplified to a more user-friendly format. These changes aim to improve the overall user experience and make the models more intuitive to work with.