llm-gemini 0.35

Simon Willison’s llm-gemini now supports Gemini 4 Argon, the new model Google made available today through the Gemini API and to customers on its $100-a-month plan. He showed off low, medium, and high pelicans, and said all three looked excellent.

That SVG obsession isn’t random. Willison says Gemini has a track record for producing strong SVGs, and that it’s the only lab he knows of whose staff have invested in SVG specifically. They’ve even bragged about their pelicans before, which is exactly the kind of niche detail that makes this release feel very Google.

The pricing is where Argon gets interesting. At launch, it’s $2 per million input tokens and $10 per million output tokens. That puts it at the same price as GPT-6 Sol, half the price of Claude Opus 5.5, and one-fifth the price of Claude Fable 5.1 and GPT-6 Astra. On December 1st, though, Argon is set to double in price, which would move it up to the same price as Opus 5.5.

Willison is careful about treating raw model prices as more than a rough clue. What matters in practice is how many tokens a model burns to get a job done inside an agentic harness. On that measure, he points to Artificial Analysis, which puts Argon’s benchmark run at $2,407. That places it near GLM-5.3 and notably below the latest models from OpenAI, Anthropic, and Meta.

Even after the price jump, Argon still looks relatively cheap by that yardstick. Willison says it would remain a little cheaper than GPT-6 Astra. In other words: the model may start as a bargain, but the real story is that Google seems to have made one that can hold its own on cost when the tokens start flying.