LLM 0.32 Adds Reasoning Traces, OpenAI Responses, and Server-Side Tools
On August 4, 2026, Simon Willison unveiled LLM 0.32, marking the most substantial enhancement since the project's inception. This update features visible reasoning traces, tools for server-side providers, revamped SQLite logs, and incorporates new models along with capabilities from the OpenAI Responses API. Enhanced versions of the llm-anthropic, llm-gemini, and llm-openrouter plugins are also part of this release. Reasoning models now provide traces to standard error, with an option to conceal them. The update is compatible with the GPT-5.6 model family, designating GPT-5.6 Luna as the default. Server-side tools from multiple providers are included, and the Python API now accommodates complete message histories, while logging enhancements are evident. Existing plugins may require updates for optimal functionality.
Key facts
- LLM 0.32 released on 4th August 2026.
- Adds support for visible reasoning traces, server-side tools, and OpenAI Responses API.
- New default model: GPT-5.6 Luna.
- New plugins: llm-anthropic, llm-gemini, llm-openrouter updated.
- New command: llm openai endpoint for one-liner prompts.
- Content-addressable message store for logging, modeled after Git.
- Python API now supports model.prompt(messages=[]) parameter.
- Tool chains can pause for human approval and resume from stored history.
Entities
Artists
- Simon Willison
Institutions
- OpenAI
- Anthropic
- LM Studio
- Datasette Agent