Google announced its latest frontier AI model, Gemini 4 Carbon, which demonstrates significant gains in cybersecurity capabilities. The model achieved the top ranking on the Artificial Analysis Intelligence Index (v5.0) benchmark. This release follows closely behind Gemini 4 Argon, which Google introduced just two weeks prior.
Gemini 4 Carbon leads its competitors, including Anthropic's Claude Opus 5.5 and OpenAI's GPT-6 Astra, on the Artificial Analysis Intelligence Index v5.0. Google highlighted Carbon's performance in cybersecurity, noting its lead in identifying vulnerabilities (SEC-bench Pro at 88.2%), exploitation (ExploitGym at 54.8%), and patching (CWE-bench at 71.0%). The model also shows improvements in coding and agentic tasks. Google's internal testing indicates Carbon reduces misalignment and fabrication rates compared to its predecessor. The company states that Carbon never successfully cheated on Terminal-Bench 4.0 and attempted to do so only 0.4 percent of the time.
Access to Gemini 4 Carbon is currently limited to a select group of cybersecurity defense companies through Google's Fairwind Program. This program provides early access to advanced AI models for trusted partners to help them proactively address cyber risks. Google has not yet announced a general public release date for Carbon. The company plans to gradually expand access to additional trusted partners before making it available to developers, businesses, and consumers.
Google has priced Gemini 4 Carbon at the same introductory rate as Argon: $2 per million input tokens and $10 per million output tokens, with cached input tokens at a 95% discount. The company has indicated that these introductory prices are temporary, with future pricing set at $4 per million input tokens and $20 per million output tokens.
The rapid release of Gemini 4 Carbon, following closely on the heels of Argon, reflects an accelerated pace in the development of frontier AI models across the industry. This competition sees AI labs vying to produce the leading model, with new announcements and benchmark claims emerging frequently.
The Fairwind Program, established in September 2026, aims to provide governments, critical infrastructure operators, and other trusted partners with Google's advanced AI and cyber defense capabilities. This initiative allows these entities early access to powerful AI models, such as Gemini 4 Carbon, to bolster their defenses against evolving cyber threats. The program emphasizes proactive cyber defense, focusing on identifying and fixing vulnerabilities before they can be exploited.
The Artificial Analysis Intelligence Index v5.0 is a composite benchmark that aggregates nine evaluations to measure AI capabilities across knowledge work, coding, science, and general knowledge. It is used to compare the performance of various AI models.