LLM 0.37 adds new models, endpoint setup and a cheaper default

Simon Willison's LLM CLI and Python library reached 0.37, adding the gpt-6.1-sol and gpt-6.1-terra models and an `llm openai endpoint add` command for OpenAI-compatible endpoints. The default model becomes gpt-6-luna at half the previous price, and embeddings now support mixed content-types.

Key points

• LLM 0.37 bundles about two months of features and fixes into one release.
• New models gpt-6.1-sol and gpt-6.1-terra are available alongside the existing roster.
• The `llm openai endpoint add` command registers models served by any OpenAI-compatible endpoint.
• The default model is now gpt-6-luna, at half the price of the previous default gpt-5.6-luna.
• Embeddings can store binary content with differing content-types, enabling models like Gemini Embedding 2.
• Fixes came from 12 contributors, nearly all AI-assisted, and the maintainer treats such PRs as bug reports with examples.

LLM, Simon Willison's CLI tool and Python library for working with language models, shipped version 0.37 with roughly two months of accumulated features and fixes. The release adds the gpt-6.1-sol and gpt-6.1-terra models, and a new `llm openai endpoint add` command that makes it simpler to register models served from any OpenAI-compatible endpoint instead of editing configuration by hand.

The default model for users who never pinned one moves to gpt-6-luna, which costs half what the previous default gpt-5.6-luna did. Embeddings handling now allows binary content with different content-types, which is required by models such as Gemini Embedding 2 that embed text, audio, images and video. Willison also notes that AI-assisted pull requests, often treated as a flood by maintainers, were manageable here: 12 contributors landed fixes, and he uses models for triage.

Real-world use

• A developer routing requests to a self-hosted OpenAI-compatible gateway instead of editing config files by hand.
• A RAG pipeline that embeds screenshots, audio clips and documents in the same store and needs correct content-types.
• A CLI-based batch job that summarizes or annotates files and wants a cheaper default model without changing flags.

Why it matters

If you script against LLM, the default model and cost profile change on upgrade, and mixed content-type embeddings widen what you can index without workarounds.

Related stories

• GitHub Copilot Fusion is generally available: Copilot routes the modelsGHGitHub Changelog: Copilot5 d ago
• Tune GitHub Copilot Inline Suggestions with Configurable EagernessVSVisual Studio Code BlogOct 26
• Backdoor found in Claude Code VS Code extension on Cursor MarketplaceHNHacker News: Claude Code3 d ago