safety benchmark
ExploitGym
ExploitGym is a large-scale, realistic benchmark built from real-world vulnerabilities across userspace programs, Google's V8 engine, and the Linux kernel. Given a vulnerability and a proof-of-vulnerability input, agents must craft a working end-to-end exploit that achieves unauthorized code execution.
Updated Aug 7, 2026
Published models3
Registry coverage3
MetricScore
EvidenceB
ExploitGym leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | OP | 33.7% | 100.0% | 3 | C | |
| 2 | OP | 23.2% | 50.0% | 3 | C | |
| 3 | OP | 12.4% | 0.0% | 3 | C |
Score distribution
Top published rows on the benchmark's original scale.
ExploitGym highlights
The top published results on this benchmark's own scale.
What is ExploitGym?
Definition and scoring fields from the benchmark registry.
ExploitGym is a large-scale, realistic benchmark built from real-world vulnerabilities across userspace programs, Google's V8 engine, and the Linux kernel. Given a vulnerability and a proof-of-vulnerability input, agents must craft a working end-to-end exploit that achieves unauthorized code execution.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- ExploitGym
- Modality
- text
- Primary category
- safety
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- exploitgym|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about ExploitGym.
Which model scores highest on ExploitGym?
GPT-5.6 Sol is currently ranked first with 33.7%.
What does ExploitGym measure?
ExploitGym is a large-scale, realistic benchmark built from real-world vulnerabilities across userspace programs, Google's V8 engine, and the Linux kernel. Given a vulnerability and a proof-of-vulnerability input, agents must craft a working end-to-end exploit that achieves unauthorized code execution.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
3 unique published model results are currently shown.
Does this benchmark affect the overall score?
This benchmark is preserved as source-native evidence but is not eligible for the current overall score.