Google launches Gemini 4 Carbon, claiming top performance on most internal benchmarks

Google announced the release of Gemini 4 Carbon, a large language model with a 1-million-token context window and multimodal capabilities for text, image, video, and speech input. The company says the model leads most of its internal benchmark evaluations, ranking first on the Artificial Analysis Intelligence Index and second on the Artificial Analysis Coding Agent Index, behind Anthropic's Claude Opus 5.5.

Pricing is set at $4 per million input tokens and $20 per million output tokens, matching the rate for the earlier Gemini 4 Argon model, with a 50 % discount on cached input tokens. For a typical Artificial Analysis Intelligence Index task, Gemini 4 Carbon costs about $7.76, roughly 23 % less than Claude Opus 5.5 and about 20 % more than GPT-6 Astra. The model also uses 123,000 output tokens per task, 29 % fewer than Claude Opus 5.5.

On security-related benchmarks, Gemini 4 Carbon achieved 88.2 % on the SEC-bench Pro vulnerability discovery test, 54.8 % on the ExploitGym exploitation test, and 71 % on the CWE-bench patching test. >"Our teams are already seeing strong gains in cyberdefense. On internal benchmarks, Carbon achieves more than 2x uplift in success rate for generating secure code over Argon. As another example, the Chrome team has reported that Carbon performed more than 10X better than another lab's large frontier model when generating high quality patches," Google wrote in a blog post.

The model is not yet publicly available and is undergoing evaluation by U.S. government agencies. Google noted that a forthcoming Anthropic release, Claude Fable 5.5, could challenge its position as the leading model once both are widely accessible.