LLM 0.32 Released with Reasoning Traces, Server-Side Tools, and OpenAI Responses Support
Simon Willison has released LLM version 0.32, introducing support for visible reasoning traces, server-side tools, and features enabled by the OpenAI Responses API. This major update also defaults to the GPT-5.6 Luna model and features redesigned SQLite logging. This release aligns the popular command-line tool with modern LLM capabilities like agentic tool-calling and reasoning models, making it easier for developers to build and debug complex AI workflows. Separating reasoning traces from standard output prevents them from breaking CLI pipes, improving developer experience. Reasoning traces are outputted to standard error (stderr) by default to avoid polluting standard output (stdout), though they can be hidden using the `-R` or `--hide-reasoning` flags. Additionally, the new `llm openai endpoint` command allows users to run one-off, unlogged prompts against any OpenAI-compatible API.
## BACKGROUND
The `llm` CLI tool and Python library, created by Simon Willison, allows developers to interact with various large language models directly from the command line. The OpenAI Responses API, introduced in early 2025, simplifies the development of agentic applications by combining chat completions with advanced tool-calling and structured outputs.