Google announces Gemini 4 Carbon with state-of-the-art cybersecurity capabilities Google

Google announced Gemini 4 Carbon, its strongest frontier model built for long-horizon engineering tasks and end-to-end cybersecurity defense.

Carbon ranks first on the Artificial Analysis Intelligence Index v5.0 and rolls out initially to select cyber defenders and trusted partners through Google's Fairwind Program, with developer and enterprise API access to follow.

What it is

• State-of-the-art cybersecurity defense. Carbon achieves 88.2% on SEC-bench Pro for detecting vulnerabilities, 54.8% on ExploitGym for exploit validation, and 71.0% on CWE-bench for automated root-cause patching.
• Long-horizon agent persistence. Engineered for sustained multi-step tasks, Carbon can navigate dead ends over extended multi-day runs to isolate and patch subtle bugs across massive codebases.
• Coding and web development gains. The model gained +89 Elo points on WebDev Arena (reaching #4 at 1768) and demonstrates over a 2x uplift in generating secure code compared to Gemini 4 Argon.
• Zero-shortcut alignment. Training mitigations reduced successful cheating on Terminal-Bench 4.0 to 0.0% while dropping high-severity misalignment alerts by nearly 80% in internal live coding environments.
• Total prompt-injection resistance. Carbon resisted 100% of evaluated attacks on the Gray Swan Indirect Prompt Injection benchmark (0.0% attack success rate).

Pricing and benchmarks

Sources

• Google DeepMind: Introducing Gemini 4 Carbon