Gemini 4 carbon: a new benchmark in cybersecurity and long-horizon AI

Google DeepMind has unveiled Gemini 4 Carbon, its latest and most advanced AI model, emphasizing state-of-the-art cybersecurity and long-horizon capabilities. The model ranks #1 on the v5.0 Artificial Analysis Intelligence Index with a score of 59, showing significant improvements in coding, agentic tasks, and cybersecurity defense. It leads in SEC-bench Pro (88.2%), ExploitGym (54.8%), and CWE-bench (71.0%). Carbon is built on Gemini 4 Argon, launched two weeks prior, and is designed for sustained agentic workflows, capable of running for days to tackle complex engineering tasks. It is priced at $2 per million input tokens and $10 per million output tokens, with cached input tokens at 95% off. The model is being rolled out gradually to cyber defenders through the Fairwind Program, with plans to expand access to developers and businesses soon. Gemini 4 Carbon excels in cybersecurity, achieving more than 2x uplift in secure code generation over Argon. The Chrome team reported Carbon outperforming another lab's model by over 10X in patch generation. Wiz, Google's Cloud security platform, also noted Carbon's leadership in web penetration testing benchmarks.