Google DeepMind announced Gemini 4 Carbon on Oct 28, and the pitch is narrow: its strongest model, built for cybersecurity work and long-running engineering tasks. It follows Gemini 4 Argon, which shipped two weeks earlier.
The announcement walks through the defense lifecycle in three steps, with a benchmark number attached to each: 88.2% on SEC-bench Pro for finding real vulnerabilities, 54.8% on ExploitGym for validating which ones can be exploited, and 71.0% on Collinear's CWE-bench for patching at the root cause. Google notes these external benchmarks were evaluated without guardrails. It also claims the top spot on the Artificial Analysis Intelligence Index v5.0 with a score of 59, plus a 0.0% attack success rate on Gray Swan's prompt-injection benchmark.
You cannot use it yet. This week a subset of cyber defenders gets access, then trusted partners through the Fairwind Program. Google says it is taking part in the U.S. government's voluntary pre-release testing process. There is no date for developers, businesses, or consumers. Introductory pricing matches Argon: $2 per million input tokens, $10 per million output tokens, cached input at 95% off.