llm 0.37
Which summary reads better? Pick one — models revealed after.Both summaries are AI-generated.
The new default model drops to half the price of the prior default, and a single `llm openai endpoint add` command now lets you wire up any OpenAI-compatible endpoint without custom config. Embeddings now carry per-item content-type metadata, which is what makes multimodal embedding models (text/audio/image/video in one index) actually work through the CLI—worth re-checking your embedding pipeline if you're mixing binary types.
The LLM 0.37 release halves default model API costs by switching to gpt-6-luna and introduces native support for multi-modal embeddings across text, audio, image, and video. This update allows production engineers to rapidly deploy multi-modal retrieval pipelines using Gemini Embedding 2 and streamline infrastructure with a simplified setup for custom OpenAI-compatible endpoints. It ultimately drives down the cost of default agent runs while removing integration friction for alternative, self-hosted inference providers.