Artificial Analysis Cyber Index v1
Indexes · 2026-09-25
Artificial Analysis Cyber Index v1 measures cyber-defense capability as the equal-weight average of three AA-run evaluations on a 0-100 scale: CWE-Bench-AA pass@1, DeepsecBench-AA F2 (median of three runs) and CyberGym-E2E-AA pass@1. All three run in Artificial Analysis' Stirrup agent harness in sandboxes without internet access, and tasks a model or provider declines on safety grounds score zero. Launched September 25, 2026 and updated October 8.
Top models (higher is better)
| Model | Score |
|---|---|
| Grok 4.7 | 56.4 |
| MiMo-V2.6-Pro | 56.1 |
| GPT-6 Luna | 52.7 |
| GLM-5.3 Flash | 49.9 |
| Mistral Large 4 Preview | 49.5 |
| Step 5 Preview | 47.7 |
| Muse Spark 1.3 | 44.3 |
| Kimi K3 | 41.3 |
| DeepSeek V4.1 Flash | 40.8 |
| GPT-6 Sol | 36.8 |
| GLM-5.3 | 36.4 |
| Beam 1.0 501B A23B | 33.7 |
| GPT-6 Astra | 33.4 |
| Gemini 3.8 Flash | 25.1 |
| MiniMax M3 | 22.4 |
| Nemotron 3 Ultra 550B A55B | 13.1 |
| Qwen3.8 27B | 12.7 |
| Qwen3.8 2.4T A95B | 12.6 |