Google announced its new frontier model, Gemini 4 Argon, on September 30. It is first rolling out to trusted cyber defenders through the Fairwind Program.
Google says a phased approach is needed and plans to expand to developers, enterprises and consumers, starting with paid API customers and Google AI Ultra subscribers. No general availability date was given.
The rollout plan is not equivalent to an available migration target.
The announced 1 million tokens is the output token limit, up from 64K before. It is meant to let the model reason deeply while generating hundreds of thousands of tokens in one trajectory, and it is not the input context size.
Introductory rates and Google-reported evaluations
Introductory pricing is USD 2 per million input tokens and USD 10 per million output tokens. Cached input tokens are priced at 95% off the input price.
After the introductory period, USD 4 input and USD 20 output apply. The announcement does not say when that period ends.
Coding and cyber defense figures such as 77.9% on DeepSWE v1.1 and 68% on CWE-bench v1 are Google’s evaluations. They were not independently reproduced for this article.
Why an output ceiling is not an availability promise
What can be used for planning now is roughly the access order and pricing terms. An actual migration comes after confirming availability, the final API specification and standard pricing. Evaluation results for the task at hand are part of that decision.
When estimating the cost of long agent runs, the output limit and input context size are handled separately. Mixing them throws off the budget.
Long runs need time and cost budgets and stopping criteria first. Allowing more output does not make generating that much output useful.
The 1M-token output limit is notable, but without an open rollout date and confirmed production terms it remains a limited-access capability.