# llm 0.37 adds OpenAI-compatible endpoint support and multi-modal embedding capabilities

> **Key takeaway:** The llm CLI and Python library update introduces better OpenAI endpoint configuration and expands embedding support for diverse content types.

**Published:** 2026-11-02T23:49:30+00:00 
**Source:** Simon Willison Web & Open Source 
**Category:** software-engineering 
**Canonical URL:** https://fosswire.org/news/llm-037-adds-openai-compatible-endpoint-support-and-multi-mo.html 

## Summary
Simon Willison has released version 0.37 of the llm tool, a command-line interface and Python library designed for interacting with large language models. This update introduces several new model options, including gpt-6.1-sol and gpt-6.1-terra, and changes the default model to gpt-6-luna to reduce costs for users.

Technically, the release improves how the tool handles embeddings by allowing different binary content types. This change is specifically designed to support advanced models like Gemini Embedding 2, which can process text, audio, images, and video. Additionally, a new command has been added to simplify the configuration of third-party models that utilize OpenAI-compatible APIs.

## Technical improvements in llm 0.37
- **API Configuration:** A new `llm openai endpoint add` command allows users to more easily integrate and configure additional models that adhere to the OpenAI API standard.
- **Embedding Versatility:** The embedding subsystem now supports varying content-types for binary data, enabling the processing of multi-modal inputs such as audio and video via models like Gemini Embedding 2.
- **Model Optimization:** The default model has been transitioned from gpt-5.6-luna to gpt-6-luna, which provides a 50% reduction in cost compared to the previous default.
- **Model Expansion:** The release adds support for new model identifiers, specifically gpt-6.1-sol and gpt-6.1-terra.

## Why it matters
This update benefits software engineers and AI researchers by lowering the barrier to using OpenAI-compatible local or third-party inference servers. The improved embedding support is critical for developers building multi-modal RAG (Retrieval-Augmented Generation) applications that require processing non-textual data.