Google previews Gemini 4 Carbon for cybersecurity, outperforming Claude Opus and GPT-6

Google has announced Gemini 4 Carbon, its latest frontier AI model, just two weeks after Gemini 4 Argon's general release. The model is initially available to selected cybersecurity defenders through Google's private Fairwind initiative, with broader access for developers, businesses, and consumers planned for a later date.

According to Google's benchmarks, Carbon scored 59 on Artificial Analysis Intelligence Index v5.0, ahead of Claude Opus 5.5 at 58, GPT-6 Astra at 57, and its predecessor Argon at 50. It also gained 89 Elo points on WebDev Arena, reaching fourth place with a score of 1769, while cybersecurity results include 88.2% on SEC-bench Pro, 54.8% on ExploitGym, and 71.0% on CWE-bench.

Google also reports improvements in safety and reliability, with internal testing showing roughly half as many unsupported claims as Argon, better handling of ambiguous requests, and greater adherence to project constraints. API pricing will initially start at $2 per million input tokens and $10 per million output tokens, with a 95% discount on cached inputs, before increasing to $4 and $20 respectively.