Gemini 4 Argon: our next era of frontier intelligence

Koray Kavukcuoglu frames Argon as Google’s next frontier model for long-horizon professional work: real-world software engineering, enterprise knowledge tasks, and cybersecurity defense. The company says it’s already in internal use and is rolling out through the Fairwind Program while Google engages the U.S. government’s voluntary pre-release model access process.

Pricing is announced up front — $2 per million input tokens and $10 per million output, with cached input at a steep discount — and the post stresses phased access and stronger guardrails before a wider developer, enterprise, and consumer release. That’s the news: capability claims plus a deliberately gated path to market.

We’re reading this as Google choosing cyber defense as both product story and safety story.

A model that can autonomously find and patch vulnerabilities is useful infrastructure if the same skills don’t become a general-purpose offense toolkit. Fairwind-first access is the right instinct after a summer of high-profile agent incidents. Voluntary government pre-release engagement is better than a silent drop; it still isn’t binding oversight.

I’m watching whether “as soon as possible” for general access outruns the monitoring Google says it’s still strengthening — especially prompt injection and misalignment checks. Internal productivity anecdotes matter less to us than whether third-party defenders can reproduce the vulnerability finds Google touts.

Useful AI earns a phased launch. We’ll celebrate Argon when independent cyber partners show durable wins — and we’ll flag every week the public still can’t stress-test the claims.

Google DeepMind company blog, September 30, 2026, announcing Gemini 4 Argon.