LLM 0.32 Released - See What Your Models Are Thinking

The latest version of LLM (Large Language Model) is here, bringing significant improvements including visible reasoning traces, server-side tools, redesigned SQLite logs, new models, and enhanced features through the OpenAI Responses API. ## Key Improvements in LLM 0.32

Reasoning Traces: Now you can see exactly how LLMs arrive at their answers with real-time reasoning displayed as standard error output (can be toggled off with -R or --hide-reasoning). This helps understand complex model behavior and debug issues.

Example of LLM showing reasoning traces

New Models: Includes built-in support for the GPT-5.6 family, with GPT-5.6 Luna as the new default model offering a balance of capability and cost.

Server-Side Tools: LLM can now utilize tools from various providers like OpenAI’s Code Interpreter (for code execution) and WebSearch - expanding its capabilities beyond simple text generation.

Enhanced Logging: Content-addressable SQLite logs provide more robust tracking and analysis of model interactions.

Additional Features

The release also includes updates to the llm-anthropic plugin with features like WebSearch, WebFetch, CodeExecution, and AnthropicMCP - allowing LLM to interact with external services through a single API call.

For those working with OpenAI endpoints, the new llm openai endpoint command enables one-line access to any compatible service without requiring configuration.