Google Launches Carbon 4

GRAND RAPIDS, MI. - Google announced the launch of Gemini 4 Carbon, its latest frontier AI model, on October 28, 2026.

The model priorities cybersecurity and complex coding tasks, building on the capabilities of its predecessor, Gemini 4 Argon, released two weeks prior.

Gemini 4 Carbon demonstrated state-of-the-art performance in cybersecurity benchmarks, achieving an 88.2% success rate on SEC-bench Pro for vulnerability discovery and a 71.0% rate on CWE-bench for patching. Google reported a more than 2x uplift in secure code generation success compared to Argon and over 10 times better performance than another large frontier model in generating high-quality patches. The model also ranked first on the Artificial Analysis Intelligence Index v5.0 with a score of 59, excelling in coding and agentic tasks.

Google implemented a phased rollout for Gemini 4 Carbon, initially providing access to a subset of cyber defenders and trusted partners through its Fairwind Program. This cautious approach stems from the model's advanced capabilities, which require careful deployment to prevent misuse. Google stated internal teams used Carbon for engineering tasks lasting several days, including diagnosing complex software bugs across large codebases.

Broader availability of Gemini 4 Carbon to developers, businesses, and consumers is planned for "as soon as possible," though no specific date is set. The introductory pricing matches Gemini 4 Argon at $2 per million input tokens and $10 per million output tokens, with cached input tokens receiving a 95% discount.