Google announced Gemini 4 Argon with limited initial access

On September 30, 2026, Google announced Gemini 4 Argon, a frontier model the company described as aimed at complex, long-running software engineering, enterprise knowledge work and cybersecurity defense. Google planned its first external access for trusted cyber defenders through the Fairwind Program.

Google’s planned access sequence

Google said the broader access phase would start with paid API customers and Google AI Ultra subscribers. Those groups follow trusted cyber defenders in the rollout plan.

The benchmark scores Google reported

Google reported four scores, each for a benchmark focused on a distinct task area.

A 1,000,000-token output limit

Google announced a maximum output of 1,000,000 tokens for Gemini 4 Argon, compared with a previous 64,000-token output limit. Output tokens are the units the model generates in its response.