Google Limits Gemini 4 Carbon Rollout to Cyber Defenders

Google will give Gemini 4 Carbon first to a subset of cyber defenders, then trusted partners, as it stages the release of its model for cybersecurity work. The company announced the model Wednesday, Oct. 28, and has not set a date for broader access by developers, businesses or consumers.

The model scored 88.2% on SEC-bench Pro for vulnerability discovery, 54.8% on ExploitGym for exploitation and 71.0% on CWE-bench for patching. The SEC-bench Pro and ExploitGym evaluations used open-source setups. SEC-bench Pro covered 183 tasks, with an average of three attempts per task.

A broader nine-test intelligence index covering knowledge work, coding, science and general knowledge listed Carbon at 59, the highest score on the index. Carbon led four of the index’s nine evaluations but did not lead in coding or knowledge work. The index measures capabilities beyond cybersecurity.

The staged release follows Gemini 4 Argon's global availability to developers and enterprises, as well as some consumer subscribers, on Oct. 13. It also builds on Google's earlier rollout of Argon to cybersecurity partners. Fairwind is an early-access program for high-priority defenders, including government agencies, health care providers and telecommunications services.

Google is participating in the U.S. government's voluntary pre-release model-testing process and plans to continue adjusting Carbon's safeguards before expanding access. Internal teams have run the model for days on complex engineering tasks. The model is designed for sustained workflows in which AI agents work through extended engineering jobs.

Google listed introductory token prices of $2 per million input tokens and $10 per million output tokens. Cached input tokens cost 95% less than the standard input rate. After the introductory period, the listed prices are $4 per million input tokens and $20 per million output tokens; Google did not specify when the introductory pricing will end.