Google announced the release of Gemini 4 Carbon, a new frontier artificial intelligence model that the company characterizes as achieving significant advances in multiple domains. The model currently ranks first on the Artificial Analysis Intelligence Index benchmark version 5.0, surpassing competing systems such as Claude Opus 5.5 and GPT-6 Astra.
The new offering represents an advancement over Argon, another Google model released earlier this month. Carbon has been engineered with particular emphasis on cybersecurity applications and is positioned as delivering substantial improvements in vulnerability detection and secure code generation. According to Google’s assessment, the model ranks first on both SEC-bench Pro, which evaluates an AI system’s capacity to identify real-world software vulnerabilities, and ExploitGym, which measures a model’s ability to convert identified vulnerabilities into functional exploits.
Google indicates that Carbon is designed to maintain focus on extended, multi-step tasks over extended timeframes, with internal testing demonstrating the model’s ability to operate continuously through failed approaches to resolve complex engineering challenges. The company also notes enhancements in prompt injection resistance, collaborative capabilities with human users, and accuracy metrics. In particular, Google reports a twofold reduction in unsupported or unsubstantiated claims relative to Argon.
Initially, Carbon will be restricted to a limited set of cybersecurity defense organizations, though Google has indicated plans to expand availability to additional partners participating in its Fairwind Program. The model will be offered at an introductory pricing structure matching Argon’s rates: $2 per million input tokens and $10 per million output tokens.
Article Attribution | Read More at Article Source
Article summary produced by Claude AI