Summary & Key Takeaways ​

LLM 0.32 introduces visible reasoning traces, showing how models "think" during execution. New server-side tools like OpenAI's CodeInterpreter and WebSearch are now supported. The default model for llm prompt is now the cost-effective GPT-5.6 Luna. Content-addressable SQLite logs have been redesigned for better data management. The llm-anthropic plugin also received substantial updates, including new Claude 5 models and server-side tools. A new llm openai endpoint command allows one-off prompts against any OpenAI-compatible API. This release significantly enhances the developer experience for interacting with LLMs locally and via various APIs.

Our Commentary ​

This is a big one for anyone wrangling LLMs locally. I'm genuinely excited about the reasoning traces; seeing the model's "thought process" directly in the terminal is a game-changer for debugging and understanding. The server-side tools integration feels like a natural evolution, pushing the boundaries of what a CLI can do with these powerful models. We've been waiting for more robust tooling in this space, and Simon consistently delivers.