Google DeepMind launches Gemini 4 Argon, its first model built for cyber defence

Google DeepMind said on 30 September 2026 that it has launched Gemini 4 Argon, a frontier AI model it is rolling out first to a group of trusted cybersecurity defenders through its Fairwind Program rather than directly to the public. TechCrunch confirmed the launch the same day, reporting that the model "was trained specifically for defensive cyber work" and can "autonomously find, validate, and patch critical software vulnerabilities."

What changed

• The model's output token limit rises to 1 million tokens, up from 64,000 for the previous version, according to Google DeepMind.
• Argon launches at an introductory price of $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 respectively once the introductory period ends, Google DeepMind said.
• Argon set a new state of the art on the DeepSWE v1.1 coding benchmark with a score of 77.9% and ranks first on Zapier's AutomationBench with a score of 51.3%, the company said.
• On CWE-bench v1, which measures a model's ability to remediate software vulnerabilities, Argon tied for first place with a score of 68%, according to Google DeepMind.
• A team of Argon agents freed up more than 300 TiB of memory across Google's data centres, with an estimated 500 TiB to 1 PiB in total savings, the company said.
• TechCrunch reported that Google's blog post claims Argon "scored significantly higher than OpenAI's GPT-6 Astra and Anthropic's Fable and Opus models across a variety of AI benchmarks."
• Security firm Wiz is already using Argon for vulnerability remediation through its Scan for Good initiative, which Google DeepMind described as "a program dedicated to protecting critical public infrastructure for free by finding and remediating high-risk exposures."

Why it matters

The staged release — to cyber defenders and internal teams first, with developer and consumer access to follow — marks a shift from the pattern of frontier labs racing straight to public launch, and comes as Google DeepMind says it is engaging with the US government's voluntary pre-release access process while it expands availability. For businesses and public bodies in the Caspian region and Türkiye that are shaping AI procurement or cybersecurity policy, Argon's narrow initial rollout for vulnerability-patching work is an early signal of how labs are managing dual-use risk before wider enterprise access.