Google Introduces Gemini 4 Carbon for Vulnerability Detection
Initially, access will be granted to a limited group of cybersecurity specialists, after which the model will be opened to developers and other trusted partners
Google has introduced Gemini 4 Carbon, a new AI model designed for complex engineering tasks and cybersecurity. It follows the Gemini 4 Argon line, released two weeks earlier. In the first phase, access will be granted to a limited group of information security specialists.
The model is designed for several stages of vulnerability handling: detecting errors, verifying their exploitability, and addressing the root cause. This approach allows not only identifying suspicious code sections but also helping developers determine which issues require prioritized fixes.
According to Google’s published results, Carbon scored 88.2% in the SEC-bench Pro vulnerability detection test, 54.8% in ExploitGym for exploitability verification, and 71% in CWE-bench for preparing fixes.
The company also reports improvements in programming and executing multi-step tasks. Based on Google’s internal assessments, the likelihood of successfully generating secure code has more than doubled compared to Argon. The number of unverified claims in tests has been reduced by approximately half.
Access will be gradually expanded through the Fairwind program. Google then plans to offer the model to developers, companies, and regular users.
For the introductory period, the rates are set at $2 per million input tokens and $10 per million output tokens. Later, prices will increase to $4 and $20, respectively.
