Simon Willison has shipped version 0.35 of llm-gemini, his plugin that connects the llm command-line tool to Google's Gemini models. The release, out on 13 October, adds support for Gemini 4 Argon, the newest model in Google's Gemini 4 line.
llm is Willison's open-source Python tool for querying large language models from a terminal, and plugins like this one extend it to cover specific providers. Installing llm-gemini with llm install llm-gemini gives access to Gemini models without writing any Google-specific code; an API key is set once with llm keys set gemini or an environment variable.
This is a small release by Willison's usual standard, adding one new model identifier rather than a feature. It follows closely on 0.34, published in September, which added gemini-3.8-flash with selectable low, medium and high thinking levels, and fixed a bug where async responses recorded the wrong model version. The release before that, 0.33, was more substantial: it moved reasoning and tool-call handling onto typed stream events, added new embedding models, and exposed Google Search, URL context and code execution as callable tools.
The pattern is familiar for anyone who follows Willison's plugin ecosystem: Google ships a model, and within days a corresponding llm release makes it reachable from the command line, often before most third-party wrappers catch up. For developers who script against multiple model providers, that quick turnaround keeps one tool current rather than several.
Source: Simon Willison
