Gemini 4 Argon hits llm-gemini at $2/million input, doubling Dec 1

llm-gemini is Simon Willison's plugin for calling Gemini models, and he shipped Gemini 4 Argon support on the day the model became available through the Gemini API and a $100/month subscription plan. He tested it with his usual pelican SVG prompts at low, medium and high settings, noting Gemini's track record of producing strong SVGs and that its staff have invested in SVG specifically — the high illustration cost him 17.166 cents.

The pricing sets up an unusual window. Argon's launch rates are the same as GPT-6 Sol's, half of Claude Opus 5.5's, and one-fifth of Claude Fable 5.1 and GPT-6 Astra. But on December 1st Argon is set to double, which would make it the same price as Opus 5.5, and still a little cheaper than GPT-6 Astra. Willison notes that per-token prices are not really comparable in agentic settings, since what matters is how many tokens a model takes to achieve a goal.

Artificial Analysis's benchmark-run cost figures offer one clue on that front: Argon's $2,407 puts it close to GLM-5.3 and significantly cheaper than OpenAI, Anthropic, and Meta's latest models. Willison also shared bonus outputs — a jaywalking armadillo in fishnet tights on Mars, a Blender-made Fabergé egg, and a DeepMind staffer's video of a pelican on a bike using fourier epicycles.