Gemini 4 Carbon tops security benchmarks, opens first to cyber defenders
Google DeepMind has launched Gemini 4 Carbon, its newest frontier model, topping the Artificial Analysis Intelligence Index v5 at 59 points while ranking fourth on WebDev Arena with an elo gain of 89 points to 1768. The company claims leading results on security benchmarks including SEC-bench Pro, ExploitGym and CWE-bench, and access is opening first to selected cyber defenders and security firms. It remains unavailable to general users and developers, with access to expand gradually. A third-party aihot test reported hallucination rising from 4% to 11% and roughly 29% fewer output tokens than Claude Opus 5.5, though the test conditions were not disclosed.
- Heat index
- 1358
- 48h sources
- 20
- 48h reports
- 31
- First seen
- 20h ago
