Google announced Gemini 4 Argon on Sept. 30, calling it its new frontier model, meaning one of the most capable AI models in existence. Few people outside Google can use it yet. Google is giving it first to a group of trusted cyber defenders through its Fairwind Program, which offers early access to groups such as governments, healthcare providers and telecom companies. Everyone else will have to wait.
The reason is Argon's skill at computer security. Google says it trained Argon for cyber defense, and that it can find, confirm and fix serious vulnerabilities, weaknesses in software that an attacker could use, on its own. In the wrong hands, Google says, the same abilities could become an equally powerful threat. So trusted defenders and Google's own teams get a version without the cyber safety limits, while the version meant for everyone else is designed to refuse requests to help with cyberattacks or attacks using chemical, biological, radiological or nuclear weapons. Google says it will also watch the notes the model writes as it reasons, along with its actions, and stop it if it starts to go beyond what a user wanted.
Google says Argon leads on long, demanding work such as coding, finance and law. On DeepSWE, a test of long, real-world programming tasks, Google says its own testing puts Argon at 77.9 percent, the best score yet. Vals AI, an independent company that tests AI models on finance, coding, legal and tax work, ranks Argon first on its overall index at 68.9 percent, just under two points ahead of Anthropic's Claude Sonnet 5.5.
Google says thousands of its employees already use the model. Google says the security company Wiz used it to find a serious flaw that could expose personal information in healthcare software used by hospitals around the world, one that earlier top models had missed. Google has not given details of the flaw.
When Argon does open up, it will go first to subscribers to Google's AI Ultra plan and to developers who pay to build it into their own apps. Prices are set per million tokens, the small chunks of text, often parts of words, that AI models read and write. Developers will pay an introductory $2 per million tokens the model reads and $10 per million it writes, rising later to $4 and $20. Google is also raising how much text Argon can produce in one go to 1 million tokens, about 750,000 words, up from 64,000 in earlier Gemini models, which it says lets the model work through hard problems in a single pass.
Google says it is taking part in a voluntary U.S. government process that gives officials access to models before release, and will open Argon to developers, businesses and the public as soon as possible once it has improved its safety limits. It has not given a date, and it has not said how long the introductory price will last.