Simon Willison shipped LLM 0.32 on August 4, 2026, a substantial update to LLM, the widely used command-line tool and Python library for talking to language models. The release adds provider-side tools, streaming reasoning traces, OpenAI Responses API support, and a Git-style logging store that stops your prompt history from ballooning.
What Happened
LLM 0.32 introduces server-side tools: capabilities that run on the provider's infrastructure rather than on your machine. That means you can now call OpenAI's CodeInterpreter and WebSearch, plus Anthropic's WebFetch and CodeExecution, straight from the CLI without wiring up your own local tool handlers.
The release also surfaces model reasoning. When a model thinks out loud, LLM 0.32 streams those traces to standard error so they stay visible without polluting piped output, and a new -R/--hide-reasoning flag turns them off when you want clean results. Under the hood, the tool is built on the current LLM codebase and ships GPT-5.6 family support with GPT-5.6 Luna as the default model.
How to Integrate This
If you script content pipelines, this release removes a chunk of boilerplate. Instead of building your own web-fetch or code-execution step, you can lean on the provider's built-in tools: pipe a research prompt through LLM with WebSearch enabled, or hand a data-cleaning task to CodeInterpreter, and let the model run the work remotely. The reasoning traces make it far easier to debug why an agent chose a given branch, since you can watch its thinking in your terminal while the final answer still pipes cleanly into the next command.
Why It Matters for Creators
LLM is the glue in a lot of solo builder and creator workflows: batch-captioning images, drafting scripts, summarizing transcripts, or chaining models inside shell scripts. Adding server-side tools turns a thin CLI wrapper into something closer to an agent runtime, without forcing you to adopt a heavier framework. For anyone who already automates with pipes and cron jobs, this is a meaningful capability jump inside a tool you already know.
Key Details
Version: LLM 0.32, released August 4, 2026.
New tools: OpenAI CodeInterpreter and WebSearch, Anthropic WebFetch and CodeExecution.
Reasoning: Traces stream to stderr by default, disable with -R/--hide-reasoning.
Logging: A content-addressable message store deduplicates repeated JSON, similar to how Git stores objects. Full notes are in the LLM changelog.
What to Do Next
Upgrade with pip install -U llm or uv tool upgrade llm, then try piping a prompt through WebSearch to see the server-side tools in action. If you maintain shell-based automations that call earlier LLM versions, review the reasoning-trace behavior before running them in production so the new stderr output does not surprise your scripts.
