Simon Willison's LLM 0.37 ships: cheaper default, any OpenAI endpoint
Two months of fixes land in the LLM CLI: a half-price default model, a command to point it at any OpenAI-compatible endpoint, and multimodal-ready embeddings.
Copy markdown
Your default model just got 50% cheaper
If you never set a default, LLM 0.37 now routes to gpt-6-luna — half the price of the previous default, gpt-5.6-luna. Upgrade the CLI and the same scripts and pipelines cost half as much, with zero code changes.
Point LLM at any OpenAI-compatible model
The new `llm openai endpoint add` command registers any OpenAI-compatible endpoint, so a local model (Ollama, vLLM, LM Studio) or a third-party gateway runs through the same CLI and Python API as the hosted ones. One interface for every model you can reach.
Embeddings go multimodal
Binary content can now carry its own content type — the piece Gemini Embedding 2 needs to embed text, audio, images, and video. Build a single vector store spanning modalities from one `llm embed` call instead of stitching separate pipelines.