Google announces Gemini 4 Argon frontier model with 1M token context and limited rollout

Gemini 4 Argon is Google's next frontier model; initial access restricted to trusted cyber defenders and internal teams while Google strengthens frontier safeguards. The model supports up to 1M output tokens and is slated to reach paid API customers and Google AI Ultra subscribers after additional testing.

In this brief: 3 sections 2 min read

- Rolling out initially to a set of trusted cyber defenders through the Fairwind Program.
- Available internally at Google and to trusted testers; broader availability planned for paid API customers and Google AI Ultra subscribers after further testing.
- Google will release Argon without cyber guardrails to trusted defenders and internal teams to enable full cybersecurity defense capabilities.
- Built for deep reasoning in long multi-step tasks across software engineering, finance, legal and cybersecurity.
- Output token limit expanded to 1,000,000 tokens (up from prior 64K limit).
- Google reports state-of-the-art scores on internal/third-party benchmarks (DeepSWE v1.1, Vals Index, LVBench, AutomationBench) per the announcement.
- Trained to autonomously find, validate, and patch critical software vulnerabilities for trusted defenders.
- Google emphasizes additional "critical frontier safeguards" including monitoring for misuse, prompt injection defenses, and alignment checks prior to wider release.
- Company is participating in the U.S. government's voluntary pre-release model access process.

Read full analysis on blog.google

Useful?