LLM 0.32 introduces reasoning traces, server-side tools, and enhanced model support

Simon Willison released LLM 0.32, the most significant update to the LLM project since its launch. The new version displays reasoning traces from reasoning-capable models to standard error output, allowing developers to inspect model thinking without affecting piped output. LLM now supports GPT-5.6 models natively with Luna as the new default for prompt commands. The update adds server-side tool execution from multiple providers, including OpenAI's code execution environment. The llm-anthropic plugin gained WebSearch, WebFetch, CodeExecution, and AnthropicMCP capabilities, enabling MCP calls within single API requests. A new OpenAI-compatible endpoint command allows one-liner prompts against any OpenAI API-compatible service without logging.