Google Says Gemini 4 Carbon Sets New Marks in Coding and Cybersecurity

Google has introduced Gemini 4 Carbon, a new frontier model it says posts top scores in areas including coding and cybersecurity. The company says Carbon builds on Gemini 4 Argon, launched two weeks earlier, and shows how quickly it is advancing its models.

According to Google's own benchmarks, Carbon reached the highest AA Index v5.0 score, a general AI benchmark, at 59 - just ahead of Claude Opus 5.5 and GPT-6 Astra. Google also reports near-sweeps in categories including agentic coding, computer use, knowledge work, science and math. These figures are vendor-reported and have not been independently verified.

Google placed particular emphasis on cybersecurity, where it says Carbon beat other models by fairly wide margins. The company says the model can find and prioritize vulnerabilities and patch them at the root cause. It also claims advances in safety, including reducing shortcuts and loophole exploitation, resisting prompt injection, and flagging unsupported claims. Google says Carbon is twice as likely to generate secure code as Argon.

As with earlier frontier models, Carbon will first go to select cybersecurity defense companies for pre-release testing. Google plans to widen access to additional trusted partners through its Fairwind Program before a broad release - the same path Argon took before reaching the Gemini app and AI search mode.

Source: Engadget