ExploitBench - release-reported exploit progress
Code
Reported progress from reaching vulnerable code through arbitrary code execution, under the ExploitBench API harness. Mean across five seeds where specified; this is distinct from binary ACE success and the raw out-of-16 capability score.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-6 Astra | 100.0 |
| GPT-5.6 Sol | 73.5 |
| Muse Wonder Full | 58.9 |
| GPT-5.6 Terra | 52.9 |
| MiMo-V2.6-Pro | 47.9 |
| GPT-5.6 Luna | 33.2 |
| MiMo-V2.6-Flash | 25.3 |