Google unveils Gemini 4 Carbon, now ranked No. 1 AI model, but only cyber defenders get it first
Google on Wednesday announced Gemini 4 Carbon, a new frontier AI model aimed at long-running coding work and cybersecurity, just two weeks after Gemini 4 Argon. It ranks first on the Artificial Analysis Intelligence Index v5.0 with 59 points, but for now only a group of vetted cyber defenders can use it, through Google's Fairwind Program.
Google on Wednesday, 28 October, announced Gemini 4 Carbon, which it calls its strongest AI model so far. The company says the model was built for long, multi-step engineering work and for cybersecurity defence. It arrives only two weeks after Gemini 4 Argon, Google's previous flagship.
Independent evaluator Artificial Analysis puts Carbon first on version 5.0 of its Intelligence Index, with a score of 59. According to VentureBeat, that places it just ahead of Anthropic's Claude Opus 5.5 (58) and OpenAI's GPT-6 Astra (57), and well above Argon (50).
Security is the main focus. Google reports 88.2% on SEC-bench Pro, a test of finding real software vulnerabilities, 54.8% on the ExploitGym exploitation benchmark and 71.0% on CWE-bench, which measures how well a model writes working fixes. Google notes that these external tests were run without its safety guardrails. On its own internal benchmarks, the company says Carbon succeeds at writing secure code more than twice as often as Argon.
Ordinary users and developers will have to wait. Google says it will first give Carbon to a subset of cyber defenders and then gradually to other trusted partners through its Fairwind Program, while it takes part in the US government's voluntary pre-release testing of AI models. Wider access for developers, businesses and consumers will come "as soon as possible", the company said, without giving a date.
Carbon will launch at the same introductory price as Argon: $2 per million input tokens and $10 per million output tokens. After the introductory period, the price rises to $4 and $20 respectively.
Outside testing also shows a trade-off. Artificial Analysis found Carbon answers more factual questions correctly than Argon (58% against 48%), but its hallucination rate on that test rose from 4% to 11%, still among the lowest of the models it measured. Google, for its part, says Carbon roughly halves unsupported claims compared with Argon in its internal testing.
Reuters described the launch as Google stepping up competition with OpenAI and Anthropic in the race to build more capable coding and cybersecurity systems.