Coding
ExploitBench Leaderboard
ExploitBench, from Carnegie Mellon University researchers Seunghyun Lee and David Brumley, scores AI agents on writing working exploits for 41 real, already-patched bugs in Chrome's V8 JavaScript engine, with modern mitigations switched on. Instead of pass/fail it grades 16 stages from crash up to arbitrary code execution, and Epoch AI reports the mean capability reached.
Source: epoch3 open models ranked+6 proprietaryData through Apr 2026
All models ranked on ExploitBench
Proprietary / closed models are shown dimmed — you can't run them locally, but they show where the open field stands.
| # | Model | Score | VRAM @ Q4 |
|---|---|---|---|
| 1 | Claude Mythos Preview Early · proprietary | 73.8% | API only |
| 2 | GPT 5.5 (unspecified) · proprietary | 47.4% | API only |
| 3 | Claude Opus 4.7 · proprietary | 26.5% | API only |
| 4 | Gemini 3.1 Pro Preview · proprietary | 26.1% | API only |
| 5 | Claude Sonnet 4.6 · proprietary | 23.6% | API only |
| 6 | Kimi K2.6 · 1026.9B | 18.4% | ~620 GB |
| 7 | GLM 5.1 · 753.9B | 18.1% | ~457 GB |
| 8 | Claude Haiku 4.5 (Oct 01, 2025, unspecified) · proprietary | 13.7% | API only |
| 9 | MiniMax M2.7 · 228.7B | 13.3% | ~138 GB |
ExploitBench: frequently asked questions
- What is the best open LLM on ExploitBench?
- Kimi K2.6 is the top open model on ExploitBench, scoring 18.4%. Among all models tested — including proprietary ones — it ranks #6. The top model overall is Claude Mythos Preview Early (Anthropic) at 73.8%.
- Can open models match proprietary models on ExploitBench?
- Not quite on ExploitBench: the strongest proprietary model (Claude Mythos Preview Early) scores 73.8%, ahead of the best open model (Kimi K2.6) at 18.4% — but you can run the open one yourself.
Scores aggregated from epoch. llmrun does not run this benchmark — see the source for methodology, or the about benchmarks for what it measures.