LLM 0.32 Release Brings Support for Reasoning Traces, Server-Side Tools, and Enhanced Logging

Simon Willison released LLM 0.32, a major update to the LLM command-line tool that adds comprehensive support for AI model reasoning capabilities and improved API integrations. The release includes visible reasoning traces from supported models displayed via standard error, server-side provider tools, content-addressable message logging, and support for the new GPT-5.6 model family. The default model for LLM is now the cost-efficient GPT-5.6 Luna, and the tool includes built-in support for OpenAI's code execution environment. The update also introduces a new LLM CLI command for executing prompts against OpenAI-compatible endpoints without logging. Accompanying this release, the llm-anthropic plugin was substantially updated with new capabilities including WebSearch, WebFetch, CodeExecution, and AnthropicMCP support. This enables LLM to execute MCP calls through Anthropic's API as part of single request-response interactions. The Python API received significant improvements to better handle complex model responses including reasoning text, tool calls, and image attachments.