Googles Gemini 4 Argon is the latest super-smart AI that most cannot get

Google launched Gemini 4 Argon on Thursday, a "frontier" AI model rolling out only to "a set of trusted cyber defenders" through the company's Fairwind Program, according to Google and reported by Mashable and 9to5Google. Koray Kavukcuoglu, SVP of Google DeepMind, wrote in a company blog post that Google is "actively engaged in the U.S. government's voluntary process for pre-release model access" while gradually expanding availability.

The limited release matters because it marks Google's first major frontier model launch since Gemini 3 Pro in November 2025, according to 9to5Google. After Gemini 3 Pro's debut, Google was briefly seen as ahead of OpenAI and Anthropic, prompting what 9to5Google described as a "code red" inside rival labs, but critics had since written DeepMind off, with one analysis firm saying "for all intents and purposes, we believe DeepMind is no longer a frontier lab."

Google says Argon excels at software engineering, enterprise legal and financial work, and cybersecurity defense, and can "autonomously find, validate, and patch critical software vulnerabilities." The model sets a new state of the art on DeepSWE v1.1 with a score of 77.9%, ranks first on Zapier's AutomationBench at 51.3%, and scores 91.7% on LVBench for long video understanding, according to Google's blog post. Google also expanded Argon's output token limit to 1M tokens, up from 64K tokens, and said a team of Argon agents freed over 300 TiB of memory by optimizing its data center usage.

After the introductory period, the price rises to $4 per 1M input tokens and $20 per 1M output tokens. 9to5Google noted the introductory rate matches OpenAI's Sol 6.1 pricing but will eventually reach the $4/$20 level Claude Opus 5.5 already charges.

Google's internal restructuring preceded the launch: an executive shuffle in August removed Demis Hassabis and Jeff Dean from DeepMind's core management team and shifted most operations from London to Mountain View, where Kavukcuoglu now reports directly to CEO Sundar Pichai, according to 9to5Google. Addressing speculation that Google might abandon frontier AI development for cheaper, product-focused work, Kavukcuoglu said there's nothing more important to Google than staying at the frontier, and that its full stack lets it optimize to reach that goal.

According to Google's own data cited by Mashable, Gemini 4 Argon outperforms OpenAI GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5 on most benchmarks. But a Bloomberg report cited by 9to5Google, based on anonymous Google employees, raised doubts about real-world performance: "While Gemini 4 has performed well on benchmarks widely used to gauge model efficacy, it does less well when employees actually put it to work," with the model struggling on certain coding tasks, according to people with direct access to the effort.

Google said cybersecurity firm Wiz is already using Argon through its Scan for Good initiative, which works to protect public infrastructure by finding and fixing high-risk security exposures for free. In an early test, the model found a critical vulnerability exposing sensitive personal data in healthcare software used by hospitals worldwide, a risk Google said earlier frontier models had missed. Google said it will keep gathering feedback from early testers before making Argon available more broadly to developers, enterprises and consumers, starting with paid API customers and Google AI Ultra subscribers.