Google said on its blog that Gemini 4 Argon, its newest model, is live worldwide. Paid-tier API users and businesses can use it in AI Studio and the Gemini platform. Ultra subscribers get it in Antigravity. Ultra or Pro subscribers can open it in the Gemini app and Search's AI Mode. Each million input tokens costs $2, and each million output tokens costs $10. Tokens are small chunks of text. Cached input is 95% cheaper.
These rates are introductory. They end on November 30, 2026, then rise to $4 and $20. Google says Argon ranks first on the Vals Index, a benchmark across finance, coding, legal and tax work. The post names no outside check. The 1M-token limit is one pool shared by input and output. Google mentions a new continuation protocol for long jobs but does not explain how it works. It sends readers to its documentation. It does not say if the cached discount continues after November 30. It gives no dates for other Gemini 4 models.
An introductory price is a trial rate, not a planning rate. Both prices double, so any job costs twice as much. A million tokens in and a million out is $12 now and $24 later. Ten million in and one million out is $30, then $60. Budget at the later rate. Since input and output share one pool, read the documentation before long jobs.