Google DeepMind launches gemini 4 carbon for cybersecurity and coding

Google DeepMind has unveiled Gemini 4 Carbon, its most advanced model to date, designed for long-term coding tasks and cybersecurity defense. The model ranks #1 on the v5.0 Artificial Analysis Intelligence Index with a score of 59. It is initially available to a select group of cyber defenders, not through a public API. Carbon is the second variant of Gemini 4, following Gemini 4 Argon, and is tailored for complex, long-running engineering projects. Internal testing showed Carbon capable of running for days on single tasks, such as debugging and patching large codebases. Carbon also excels in full-stack web development and multimodal understanding, achieving 1769 on WebDev Arena. It outperforms several competitors in specific benchmarks, including Terminal-Bench 4.0 and SWE-sweep. Google claims Carbon is over 10 times better than another lab’s model at producing high-quality patches, according to the Chrome team. The model is part of the Fairwind Program, which prioritizes government and critical infrastructure partners. Pricing remains at $2 per million input tokens and $10 per million output tokens during the introductory period.