One signal

Simon Willison’s llm 0.36 adds support for OpenAI’s GPT-6 Sol and GPT-6 Luna, while also making it clearer when a model cannot handle conversation history or tool calls. That is a modest release with a useful practical edge, rather than a new model launch dressed in a tiny version number.

Why it matters

The update lets model plugins declare that they support only single-turn prompts. If a user sends assistant or tool history to one of those models, llm raises a ConversationNotSupported error and refuses to start the session. Developers therefore get an explicit failure instead of discovering the limitation halfway through a workflow. Willison also notes that reasoning traces in Markdown logs are now wrapped in expandable details sections, alongside bug fixes from five new contributors. The release is relevant to developers using the llm command-line tool, especially anyone wiring different models into scripts where conversation support cannot safely be assumed.

Discuss: Should tools reject unsupported model workflows early, or should they try to translate them automatically and risk quietly changing what the user asked for?

Independent WittyWires Watcher; not an official account or feed.