A release with a narrow front door

Google announced Gemini 4 Argon on September 30. For now, the model is going to a selected group of cyber defenders through its Fairwind Program. Google's public announcement says wider access for developers, enterprises and consumers will follow after more testing, starting with paid API customers and Google AI Ultra subscribers. It does not give a date for that wider release.

That distinction matters. Argon is a real model with early users, but a developer reading the launch page cannot assume it is generally available to build on today. Fairwind says participating organisations are vetted and that access is restricted to approved defensive and research uses.

The published price has an expiry, though not a date

Google lists an introductory price of $2 per million input tokens and $10 per million output tokens. Cached input is priced 95% below the input rate. A footnote says the standard input and output rates will become $4 and $20 per million after the introductory period. The announcement does not say when that period ends.

Those figures are useful for planning, but not a finished cost estimate. A real job can use different amounts of input, output and cached text, and may need several attempts or a person to review it. Teams should wait for the actual access terms and measure complete work before comparing bills.

The benchmark table is not a single verdict

Google reports strong results on coding, business and cyber evaluations, including 77.9% on DeepSWE v1.1 and a 68% tie on CWE-bench v1. Those are published results from Google. Its own comparison table also shows other models ahead on some tests, so it does not support a blanket claim that Argon leads everywhere.

The first deployment focuses on security because the same ability to find and patch vulnerabilities can also be misused. Google says approved Fairwind partners can use a version without some cyber guardrails, under access controls and restrictions on malicious tasks. How that arrangement performs in day-to-day defensive work, and how the safeguards will change for wider users, remain open questions.

Sources

  1. Google: Gemini 4 Argon announcement, September 30Primary dated announcement for staged availability, introductory and later prices, stated evaluations and planned broad rollout.
  2. Google DeepMind: Gemini model comparisonPrimary published evaluation table, including tests on which competing models score higher.
  3. Google DeepMind: Fairwind ProgramPrimary description of the vetted early-access group, permitted uses and governance restrictions.