Benchmarks / Agentic
ExploitBench v0.1
Agentic memory-corruption exploitation on real V8 CVEs; mean share of the 16-flag capability ladder (coverage → arbitrary read/write → code execution) across seeds.
ExploitBench v0.1
Higher is better · best effort per model
full table below
| # | Model | Effort | Score |
|---|---|---|---|
| 1 | Mythos Preview | – | 78.0% |
| 2 | GPT-5.5 Codex | xhigh | 72.0% |
| 3 | GPT-5.5 | xhigh | 47.0% |
| 4 | Opus 4.7 | max | 28.0% |