Google announced Gemini 4 Argon on September 30, 2026, and began rolling it out to a small group of trusted cybersecurity defenders. The model is not generally available. Google says it will widen access after further testing.

Developers, businesses and everyday users will have to wait. Google says broader access is planned for paid application programming interface customers and Google AI Ultra subscribers, but has not announced a date. Google’s announcement and Axios’s report both describe a restricted rollout.

What Google says Argon can do

Google designed Argon for complex, long-running work in software engineering, legal and financial research, and cybersecurity defense. The company says it can find, validate and patch software vulnerabilities.

Google also says Argon can generate up to one million tokens in a single run. That allows for much longer outputs, but a high output limit alone does not show whether a model can complete a task accurately.

What the 68% score measures

Google says Argon scored 68% on CWE-bench v1, a benchmark for testing how well models remediate security vulnerabilities. The company says the result ties for first place on that test.

That score is limited to one benchmark. It does not measure performance across legal or financial research, general coding, or real-world security operations. Broader testing will be needed to assess those uses.

Who gets access next

Google is using its Fairwind Program to give initial access to trusted cyber defenders. The company says it will use their feedback to strengthen safeguards before expanding access to developers, businesses and consumers.

  • Current access: a small group of trusted cybersecurity defenders.
  • Planned next access: paid application programming interface customers and Google AI Ultra subscribers.
  • Wider release date: not announced.
Argon’s 68% score is a result on one vulnerability-remediation benchmark, not proof of broad performance. For most users, access remains limited and the wider release date is unknown.