WMDP-Cyber
Cybersecurity · 2024-03-05
WMDP-Cyber is the cybersecurity split of the Weapons of Mass Destruction Proxy benchmark: four-option multiple-choice questions that proxy for hazardous offensive-cyber knowledge. Scores report accuracy; higher means more dual-use knowledge.
Top models (higher is better)
| Model | Score |
|---|---|
| Grok 4.20 | 91.0 |
| Grok 4.1 | 84.0 |
| Grok 4 Fast | 81.4 |
| Grok 4 | 79.0 |
| Grok Code Fast 1 | 62.1 |