Google DeepMind

Artificial Intelligence · Cybersecurity

Launch

Google launches Gemini 4 Carbon, its top model for cyberdefense

Early access is being rationed to vetted defenders because the same vulnerability-finding skill can be weaponized just as easily.

• Google DeepMind launched Gemini 4 Carbon on Oct. 28, 2026, calling it its strongest model yet, built for cybersecurity defense and long-horizon agentic coding, with staged access starting via the Fairwind Program.
• Carbon arrives just two weeks after Gemini 4 Argon, underscoring how fast Google DeepMind is now shipping frontier models in its Gemini family.
• Carbon tops the Artificial Analysis Intelligence Index v5.0 at 59 and leads cybersecurity benchmarks: 88.2% on SEC-bench Pro for finding bugs, 54.8% on ExploitGym for exploitation, and 71.0% on CWE-bench for patching.
• Google says Carbon writes secure code at more than twice Argon's success rate internally, and its Chrome team found Carbon over 10 times better at generating high-quality patches than a rival lab's frontier model.
• Carbon is priced the same as Argon at $2 per million input tokens and $10 per million output tokens, but it's currently limited to a small group of cyber defenders ahead of wider rollout.
• Google says it's engaged in the U.S. government's voluntary pre-release testing process, and its Fairwind early-access program now spans more than 650 organizations in government and critical infrastructure.
• Google's staged, defender-first rollout reflects an industry-wide reckoning that AI capable of finding and patching software flaws is equally capable of exploiting them, turning cyberdefense into frontier labs' newest competitive battleground.