Google began rolling out Gemini 4 Argon on September 30, restricting its next-generation frontier model to vetted cybersecurity defenders before broader availability1,2. The company said the model is state of the art on certain coding and knowledge-work benchmarks, outperforming OpenAI's frontier model GPT-6 Astra on several of them.
Outside Google's own teams, only members of its Fairwind Program can access Argon for now. The Fairwind Program opened on September 3. Google said it will expand access after additional testing, with paying subscribers first in line.
Built for long-running tasks
Google said Gemini 4 Argon is built to handle complex, long-running tasks in software engineering, finance, legal work, and cybersecurity. "Argon is a well-rounded model that has frontier capabilities across several domains," said Tulsee Doshi, head of Gemini products at Google DeepMind. Doshi added that Googlers have been putting the model through its paces in recent weeks, "with many relying on it for their hardest coding and research problems".
Google is also participating in the U.S. government's voluntary pre-release process for the model.
A long gap at the top
Gemini 4 arrives after a prolonged stretch in which Google kept shipping smaller, cheaper Flash models while OpenAI and Anthropic pushed forward with their own frontier releases. Axios reported that in May, Google CEO Sundar Pichai said the company's next major model release would come in June; that timeline slipped. Gemini 4 comes out nearly a year after the previous flagship generation, Gemini 3.
ANALYSIS The cybersecurity-first rollout through the Fairwind Program is a narrower launch strategy than the broad developer previews that have become standard for frontier models. Pairing the restricted release with participation in the U.S. government's voluntary pre-release process frames the rollout around safety credibility rather than developer reach.
Google's decision to benchmark Gemini 4 Argon against GPT-6 Astra positions the model directly against OpenAI's current flagship, though independent third-party evaluations are not yet available from the evidence at hand.
The limited initial access means most developers and enterprise customers will wait for expanded availability, the timeline for which Google has not specified beyond "after additional testing".