- Published on
New release of LLM adds support for reasoning traces, OpenAI Responses, server-side tools, and smarter logging
- Authors
- Name
- DeepHorus
On August 4, 2026, developer Simon Willison released LLM 0.32, calling it the most significant version since the command-line tool's initial launch. The update introduces visible reasoning traces, server-side provider tools, and a redesigned content-addressable SQLite logging system. It also adds support for new models and features powered by the OpenAI Responses API.
LLM is an open-source command-line utility for interacting with large language models. Previously, the Python API required creating a conversation and sending messages one at a time, an abstraction that hindered advanced use cases. This release removes that limitation by supporting full message histories and streaming events. The companion llm-anthropic plugin also received substantial updates.
For CLI users, reasoning models now display their thinking traces to standard error, with a -R/--hide-reasoning flag to turn them off. LLM 0.32 includes built-in support for the GPT-5.6 family, making GPT-5.6 Luna the new default model. Server-side tools such as OpenAI's CodeInterpreter and WebSearch can now be invoked directly from the command line.
The updated llm-anthropic plugin adds WebSearch, WebFetch, CodeExecution, and AnthropicMCP, allowing single request/response interactions with MCP servers. A new llm openai endpoint command enables one-off prompts against any OpenAI-compatible endpoint without logging. For instance, users can run a local Gemma 4 12B model via LM Studio and uvx with no installation.
In the Python API, developers can now pass a full list of messages via model.prompt(messages=[]). Streaming events have been redesigned to handle reasoning text, output strings, tool calls, and image attachments. These changes make it possible to build a robust implementation of the OpenAI chat completions API, released as the llm-chat-completions-server plugin.
This release reflects a response to the unpredictable evolution of AI models. By embracing server-side tools and MCP, LLM aligns with the growing trend of tool-using AI systems. The new architecture simplifies complex workflows and future-proofs the tool. Looking ahead, Willison is likely to continue expanding support for Responses API features and plugin integrations.
