adtestbench

AI news · Google

Google releases Gemini 4 Argon to cyber defenders first

Google’s new frontier model goes to security partners on day one, with paid API customers and Ultra subscribers next.

By Alexander Bleu , 02:00 UTC

Google released Gemini 4 Argon, its new frontier model, on Wednesday 30 September 2026. The first users are security teams in the Fairwind Program, Google says. Google says paid API customers and Google AI Ultra subscribers come next, and it has set no date.

Retro-futurist illustration: a striped sunset over a grid horizon under a starry sky, with a rocket standing on the horizon.
Drawn by adtestbench from Gemini 4 Argon,

The model is built for long reasoning jobs, with defensive cyber work as its headline use. Google says Argon takes a serious software flaw from discovery to a working fix without a human step. Inside Google, engineers already run it on bug hunts and codebase migrations, TechCrunch reports.

Google’s own figures put Argon first or tied for first on 13 of the 18 benchmarks it published. It scores 77.9 per cent on the DeepSWE coding benchmark against 74.2 per cent for Claude Opus 5.5 and 74.1 per cent for GPT-6 Astra, VentureBeat reports. One answer can now run to a million tokens, up from 64,000.

Introductory prices are $2 per million input tokens and $10 per million output tokens. They later double to $4 and $20, Google says.

For teams that buy model access, the launch price matches Claude Sonnet 5.5’s rate card. The catch is access: most buyers cannot call Argon today.