Google ships Gemini 4 Carbon, built for cyber defence first
Google's newest frontier model tops its own benchmarks for finding, exploiting and patching vulnerabilities, but only a handful of cyber defenders can use it so far.
Google released Gemini 4 Carbon on 28 October, its strongest model yet for cybersecurity and sustained agentic work, two weeks after Gemini 4 Argon. Google says it ranks first on the Artificial Analysis Intelligence Index v5.0, with a score of 59, and roughly doubles secure code generation over Argon.
On Google’s own benchmarks, Carbon scores 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploiting them and 71% on CWE-bench for patching them. It also blocked every prompt injection attempt in Google’s own testing, against a 0.7% success rate for Argon. None of that is generally available yet: Carbon ships this week only to a subset of cyber defenders, with wider access phased in through Google’s Fairwind Program and full availability promised “as soon as possible.” Introductory API pricing is $2 per million input tokens and $10 per million output, rising to $4 and $20 once that period ends.
Why it matters: if you build or evaluate security tooling, this sets the benchmark to measure against and the price to plan around, once it actually ships to you. Until then, there is nothing to test.
The caveat: every number here is Google’s own claim, and the model itself is gated behind a vetting programme, so none of it is independently verified yet.