Google releases Gemini 4 Carbon: a frontier model for cybersecurity

Gemini 4 Carbon is Google's new frontier model with top-tier cybersecurity and long-horizon task capabilities. On the Artificial Analysis Intelligence Index v5.0 it ranks first with a score of 59 and leads key security benchmarks: SEC-bench Pro (vulnerability discovery) 88.2%, ExploitGym (exploitability verification) 54.8%, and CWE-bench (fixes) 71.0%.

The model builds on Gemini 4 Argon released two weeks ago and shows the pace at which Google is pushing boundaries. Internal teams report Carbon runs for days at a time and works through dead ends on complex engineering tasks — such as finding and fixing subtle bugs in large codebases.

Google deliberately chose a cautious rollout and joined the voluntary US pre-release testing process. Starting this week, Carbon will be available to a handful of cyber defenders, with more trusted partners joining through the Fairwind program. Feedback is collected before the model reaches developers, enterprises, and consumers.

A notable detail from the security report: Carbon reduced successful cheating on Terminal-Bench 4.0 to 0.0%, with cheating attempts dropping to 0.4% — while overall accuracy improved. Prompt-injection resistance on the Gray Swan benchmark is 100%.

Introductory pricing is $2 per million input tokens and $10 per million output tokens (same as Argon), with cached inputs discounted 95%. After the introductory period, price rises to $4/$20.