Google released Gemini 4 Argon Tuesday, pitching it as its most powerful model yet for software engineering, enterprise knowledge work and cybersecurity defense, according to the company's announcement and TechCrunch.
Argon can autonomously identify, validate and patch software vulnerabilities and holds a 1 million-token context window for long, multistep tasks, Google said. The model scored 77.9% on the DeepSWE v1.1 coding benchmark and tied for first at 68% on CWE-Bench v1, a vulnerability remediation test, according to the company.
General availability is on hold. Google is rolling out Argon first to a vetted group of government and corporate security teams through its Fairwind Program, and the company said it is already using the model internally, according to The Verge and TechCrunch. Google said it would ship the model to that trusted group without some of its cyber guardrails so they can use its full defensive capability.
Google said Argon outperforms OpenAI's GPT-6 Astra and Anthropic's models on independent benchmarks compiled by Vals, TechCrunch reported. "Frontier performance in complex workflows across real-world software engineering, enterprise knowledge work like legal and finance, and cybersecurity defense" is how Koray Kavukcuoglu, chief AI architect at Google DeepMind, described the model, according to The Verge.
Gating a flagship model behind a cybersecurity vetting program is new territory for a consumer-facing lab. For builders, the strongest coding and vulnerability-patching assistant on the market right now is only available if you can get into the program, a bet that defensive use needs to go out the door before offensive use follows.