Google's new Gemini 4 Argon AI ships to cyber agencies first, not the public

Google unveiled Gemini 4 Argon on September 30 as a new top-tier AI model aimed at cybersecurity, coding, and enterprise knowledge work. The model is not publicly available. Google's Fairwind program is shipping it first to vetted governments and cyber authorities; broader developer, enterprise, and consumer access is forthcoming, with no firm date in Google's announcement.

The headline benchmark result is a tie, not a clear win. Google says Argon scored 77.9% on DeepSWE v1.1, a public test of how well models find and patch security flaws, matching the score Business Insider attributes to OpenAI's GPT-6 Astra. Google also raised Argon's output cap to 1 million tokens, up from 64,000.

Google framed the launch as evidence it can rejoin the AI frontier after a year of trailing OpenAI and Anthropic. The company said Argon agents analyzed fleet-wide profiling telemetry and freed more than 300 tebibytes of memory, roughly 330 trillion bytes, without new hardware. In a separate example, agents replaced 32,000 lines of SIMD code in an existing Rust video codec and ran it 2.7 times faster than the prior port.

That comeback framing carries a hedge. Google previously promised a Gemini 3.5 Pro release in June that was postponed internally several times and ultimately not released, Business Insider reports. Whether Argon holds the frontier remains an open question until independent testing on matched conditions is published.