Google has unveiled Gemini 4 Argon, calling it its most capable AI model yet, with a particular focus on defensive cybersecurity. The model is designed to autonomously find, validate, and patch software vulnerabilities, and Google says it is already being used internally for coding and large-scale codebase migrations. But unlike previous broad consumer launches, Argon is initially available only to a select group of vetted partners through Google's Fairwind Program.
The restricted rollout reflects safety concerns. The Verge reports Google is 'actively engaged' in the U.S. government's voluntary pre-release model access process and will strengthen frontier safeguards before wider availability. Ars Technica notes the company has built in chain-of-thought monitoring to stop misalignment. The launch comes a day after OpenAI's DevDay and amid an escalating race, with Google claiming Argon outperforms rivals like GPT-6 Astra and Anthropic's Fable on benchmarks, though those claims are not independently verified.
Pricing and access details: Ars Technica reports API pricing at $2 per million input tokens and $10 per million output tokens, with a 1 million token output limit, up from 64,000. General availability will start with paid API users and Google AI Ultra subscribers after cybersecurity testing. CNBC adds that the release follows CEO Sundar Pichai signing a voluntary White House agreement on AI safety. While sources agree on the core announcement, they differ on emphasis: TechCrunch and CNBC frame it as a competitive move, while The Verge and Ars Technica highlight the cautious, safety-first approach.