Google has unveiled Gemini 4 Argon, its most powerful AI model yet. It announced the model on September 30. But almost no one can use it yet.

Argon is rolling out first to trusted cyber defenders through Google’s Fairwind Program. Defenders get it “without cyber guardrails”. Google says the model can autonomously find, validate and patch critical software vulnerabilities.

Google trained Argon for cybersecurity defence. It says the model delivers frontier performance in software engineering and knowledge work. The output limit is 1 million tokens, up from 64,000.

On Google’s numbers, Argon scores 77.9 per cent on DeepSWE v1.1. That is ahead of Claude Opus 5.5 at 74.2 per cent and GPT-6 Astra at 74.1 per cent. Independent tests are cooler. Artificial Analysis scores it 53, tied with GPT-6 Astra.

The introductory price is $2 per million input tokens and $10 per million output tokens.

Google has put Argon to work internally. It migrated more than 800,000 lines of the Fuchsia Zircon kernel from C and C++ to Rust. It freed over 300 TiB of memory across its data centres.

Wider availability will follow in stages after more work on guardrails.