Version 0.37 of the llm CLI tool and Python library has been released, representing roughly two months of accumulated features and fixes. The update adds several convenience and capability improvements aimed at simplifying model configuration and handling multimodal embeddings.
What's in this release?
-
New models: the release registers two new models, gpt-6.1-sol and gpt-6.1-terra, giving users additional model choices within llm.
-
Easier OpenAI-compatible endpoint configuration: a new command, llm openai endpoint add, was introduced to simplify adding further models that use OpenAI-compatible endpoints.
-
Default model switch: if a user has not set a custom default, llm now uses gpt-6-luna as the default model. According to the maintainer, this new default costs about half as much as the previous default, gpt-5.6-luna.
-
Multi-type embedding support: llm's embedding support now accepts that different binary contents may carry different content-type values. This is necessary for models that embed multiple modalities (text, audio, images, video); the maintainer cites Gemini Embedding 2 as a concrete example.
AI-assisted contributions and maintainer experience
Open source maintainers frequently complain about a flood of AI-assisted pull requests; llm has been notably affected. The project maintainer, however, does not view this as a problem. They used Opus 5.5 and GPT-6 Astra to assist with triage and to identify the most useful fixes among incoming contributions.
This release includes fixes from 12 different authors, and the majority of those contributions were AI-assisted. The maintainer describes many incoming PRs as bug reports accompanied by examples: sometimes they accept the submitted change, and sometimes they implement their own fix based on the report.
Contributors
Thanks to the following people for contributing to or spending tokens on this release: 0d00ciallo0721, ac1982, Allen Flux, Bian Yuhan, Bishop Kammeraad, Chen Yufeiyang, Christopher Krah, Ethan Stoner, Jiachen Xie, Olamide Faniyan, Sharat Chandra Sai Boddu, and nosmile99.
Why this matters
The changes make it simpler to configure OpenAI-compatible endpoints, lower costs for users who rely on the default model by switching to a cheaper default, and prepare llm for embeddings of multimodal content. These are practical, incremental improvements that ease integration with different models and media types.
Summary
llm 0.37 is a modest but useful update: it brings new model registrations, a helper command for endpoint configuration, a cost-saving default model change, and extended embedding support for multimodal inputs. The release also demonstrates that AI-assisted contributions can be a manageable and valuable part of open source maintenance.
