Google Releases Its New Gemini Model to Cyber Defenders Before the Public
Google announced a new AI model, Gemini 4 Argon, on Sept. 30 and is releasing it first to a limited set of outside cybersecurity defenders through what it calls the Fairwind Program, rather than to the public.
In its announcement, the company says capabilities at this level require a phased release, and that it will keep gathering feedback from early testers before opening Argon to developers, enterprises and consumers. Google says it is also taking part in a U.S. government voluntary process for pre-release access to models.
Those outside defenders and Google's own internal teams will get a version released without its cyber guardrails, the company says, so they can use what it describes as the model's full defensive capability. It names one user, the security company Wiz, and says Argon found a critical flaw in healthcare software used by hospitals that earlier models had missed. The company says the model is already running on engineering work inside Google.
On the company's own figures, Argon can produce up to 1 million output tokens in a single run, against 64,000 for the model it follows. Tokens are the chunks of text a model writes. Argon will cost $2 per million input tokens and $10 per million output tokens at an introductory price that later doubles.
According to Google, Argon reaches 68% on CWE-bench v1, a test of whether a model can fix security flaws, tying for first on the benchmark. The announcement was written by Koray Kavukcuoglu, Google DeepMind's SVP and chief AI architect. Broad access, Google says, will begin with paid API customers and Google AI Ultra subscribers.
