Google Unveils Gemini 4 Carbon, Tops AI Benchmarks in Cybersecurity

Google has released Gemini 4 Carbon, its newest frontier model, which it says delivers a “significant leap” in coding, agentic tasks and cybersecurity. The model currently tops the Artificial Analysis Intelligence Index (v5.0) benchmark, beating Claude Opus 5.5 and GPT‑6 Astra. Gemini 4 Carbon also holds the number‑one spot on SEC‑bench Pro, a test of real‑world bug discovery, and on ExploitGym, which measures a model’s ability to turn vulnerabilities into exploits. It is presently available only to a select group of cybersecurity defense firms but will expand to other partners through Google’s Fairwind Program. Google claims the model can run for days on complex, multi‑step engineering tasks, better avoids shortcuts and prompt‑injection attacks, and offers greater transparency by cutting unsupported claims in half compared to its predecessor Argon. Pricing will mirror Argon’s $2 per million input‑token and $10 per million output‑token rates.