Google unveiled Gemini 4 Carbon on Wednesday, its latest frontier model that it claims outperforms competing models from Anthropic and OpenAI in cybersecurity.
"Gemini 4 Carbon is our strongest model with state-of-the-art cybersecurity and long horizon capabilities," Google said in a blog post.
Carbon is the next iteration following the release of Gemini 4 Argon two weeks ago. Google says its teams have seen Carbon work "for days" on complex engineering tasks and finding and fixing bugs in large codebases.
According to Google's published benchmarks, Carbon beat OpenAI's GPT-6 Astra and Claude Opus 5.5 on ExploitGym, which tests vulnerability exploits, and CWE-bench, which tests a model's ability to patch security vulnerabilities.
However, coding test results came in a bit more mixed. Carbon led its rivals on SWE-sweep, which tests finding and fixing bugs in codebases, and Terminal-Bench 4.0. It came in behind both on FrontierSWE 2, which measures performance on engineering and research tasks.
Google also says Carbon ranked first on the latest Artificial Analysis Intelligence Index with a score of 59, compared with a 58 for Opus 5.5 and 57 for GPT-6 Astra.
Gemini 4 Carbon is currently only available to select cyber defense experts before expanding to other partners through its Fairwind Program.
After it takes in feedback from the first phase, it will make Carbon available to other customers. It will launch at the same price as Argon: $2 per million input tokens and $10 per million output tokens.