llm 0.37

What it claims

LLM 0.37, released 2nd November 2026, is a release of Simon Willison's LLM CLI tool and Python library that collects about two months of features and fixes. The default model changes from gpt-5.6-luna to gpt-6-luna, which Willison says is half the price. The default changes only for users who have not set their own. The release adds two OpenAI models, gpt-6.1-sol and gpt-6.1-terra. It also adds a command for configuring further models served through OpenAI-compatible endpoints.

How it works

Embedding support now treats binary inputs as possibly having different content types within the same model. Willison says this change is needed for models such as Gemini Embedding 2, which embeds text, audio, images and video.

The release includes fixes from 12 contributors, and Willison says almost all of them were AI-assisted. He uses Opus 5.5 or GPT-6 Astra to triage incoming pull requests and pick out the most useful fixes. He treats each PR as a bug report with an attached example: sometimes he merges it, and sometimes he writes his own fix. He reports that the review burden was much smaller than he expected.

How to use it

The new endpoint command is named on the page without arguments or flags:

llm openai endpoint add

If a script or workflow relies on the implicit default model, it will run on gpt-6-luna after upgrading. To keep the old behaviour, set a default explicitly.

Limits

The post is a short release announcement. It gives no exact prices, no command syntax beyond the command name, no API details for the content-type change, and no migration notes. It does not link a changelog in the fetched text, and it does not say whether existing embedding code or stored embeddings need changes.