Cybersecurity knowledge assessment and autonomous exploitation on PACEbench v2.0 (test)
59A-CVE Success RateClaude Sonnet 4.5
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Claude Sonnet 4.5Mode=Thinking2026.02 | 59 | 37 | 13.3 | 33.3 | 33.5 | |
| Gemini-3-Pro2026.02 | 47 | 16 | 6.7 | 0 | 16.1 | |
| GLM-4.72026.02 | 41 | 21 | 6.7 | 0 | 16.6 | |
| GPT-5.2-2025-12-11Release Version=2025-12-112026.02 | 41 | 37 | 6.7 | 33.3 | 28 | |
| MiniMax-M2.12026.02 | 35 | 5 | 0 | 33.3 | 15.3 | |
| Qwen3-max2026.02 | 35 | 26 | 13.3 | 0 | 19 | |
| Doubao-seed-1-8-2512282026.02 | 35 | 26 | 6.7 | 0 | 17 | |
| Seed-OSS-36B-Instruct2026.02 | 29 | 5 | 0 | 0 | 7.5 | |
| Kimi-K2-Instruct-09052026.02 | 24 | 5 | 0 | 0 | 6.3 | |
| Grok-42026.02 | 6 | 0 | 0 | 0 | 1.2 |