Glonce

LLM 0.37 adds gpt-6.1-sol and gpt-6.1-terra, switches default to gpt-6-luna

toolsThe Glonce editorsSimon Willison🔥49Glonce rating49Impact35Novelty35Relevance90Surprise25Players30Credibility78

Simon Willison has released a new version of his LLM command-line tool and Python library, listed as llm 0.37. He describes it as a big one, representing around two months of accumulated features and fixes.

The release adds the models gpt-6.1-sol and gpt-6.1-terra. It also brings a new command, llm openai endpoint add, which makes it easier to configure additional models that use OpenAI-compatible endpoints.

The default model changes. For anyone who has not set their own default, LLM now uses gpt-6-luna, which Willison says costs half the price of the previous default, gpt-5.6-luna. No absolute prices are given.

Embeddings support has also been widened. It now accepts that different binary content may have different content-types. Willison says this is necessary for models like Gemini Embedding 2, which can embed text, audio, images, and video.

The second half of the post is about contributions. Willison notes that many open source maintainers complain about a flood of AI-assisted pull requests. LLM, he says, is the project of his that has been most affected, and he is not finding it a problem. The main reason: he has Opus 5.5 or GPT-6 Astra help him with triage, to spot the most useful fixes.

This release includes fixes from 12 authors, almost all of them AI-assisted, and reviewing them was not nearly the burden he expected. He treats such pull requests as bug reports with attached examples: sometimes he lands them, sometimes he implements his own fix. He closes by thanking the 12 for helping out with (or spending tokens on) the 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.

Key facts

  • LLM, Simon Willison's CLI tool and Python library, ships a release covering around two months of accumulated features and fixes.
  • New models gpt-6.1-sol and gpt-6.1-terra arrive, along with an llm openai endpoint add command for OpenAI-compatible endpoints.
  • The default model for users without their own setting is now gpt-6-luna, which Willison says is half the price of the previous default, gpt-5.6-luna.
  • Embeddings now accept different content-types for binary content, which Willison says models like Gemini Embedding 2 (text, audio, images, video) need.
  • The release includes fixes from 12 authors, almost all AI-assisted; Willison says reviewing them was far lighter than expected, helped by Opus 5.5 or GPT-6 Astra for triage.

Why it matters

Two things stand out. The default model moves to gpt-6-luna at half the price of gpt-5.6-luna, so users who rely on the default may see lower costs (the source gives no absolute prices). And Willison offers a concrete counterexample to the complaint that AI-assisted pull requests swamp maintainers: in his most affected project, he says the review burden was much lighter than he expected.

Who it affects

Users of the LLM command-line tool and Python library, especially those who have not set their own default model, since they are switched to gpt-6-luna. Anyone who wants to add models served through OpenAI-compatible endpoints gets the new llm openai endpoint add command. People embedding non-text content with models like Gemini Embedding 2 are affected by the embeddings change. Open source maintainers dealing with AI-assisted pull requests may find his workflow relevant.

How to use it

Users with no default of their own need do nothing to get gpt-6-luna as the default. To use the new gpt-6.1-sol and gpt-6.1-terra models, they are now part of the release. To configure extra models behind OpenAI-compatible endpoints, run llm openai endpoint add. For embeddings, binary inputs can now carry different content-types, which is what multimodal embedding models need. The source gives no installation or upgrade instructions and no changelog link in the provided text.

How solid is it

This is a first-party announcement by the tool's author, written in the first person. The product changes are stated plainly. The cost claim is relative only: half the price, with no figures. The claims about the review burden and the triage help are Willison's own account of his experience, with no counts of pull requests received or merged.

Risks and caveats

The post does not say what gpt-6.1-sol, gpt-6.1-terra or gpt-6-luna are beyond their names. It does not describe the individual bug fixes from the 12 authors, how the triage with Opus 5.5 or GPT-6 Astra works, or what share of fixes were not AI-assisted beyond 'almost all'. Changing the default model affects anyone relying on it, and Willison's comfortable experience with AI-assisted pull requests is one maintainer's report on one project.

“I really see them as bug reports with attached examples - sometimes I land them, sometimes I implement my own fix.”

— Simon Willison