safety benchmark
CyberGym
CyberGym is a benchmark for evaluating AI agents on cybersecurity tasks, testing their ability to identify vulnerabilities, perform security analysis, and complete security-related challenges in a controlled environment.
Updated Aug 7, 2026
Published models11
Registry coverage11
MetricScore
EvidenceB
CyberGym leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | GO | 83.2% | 100.0% | 11 | C | |
| 2 | AN | 83.1% | 90.0% | 11 | C | |
| 3 | OP | 81.8% | 80.0% | 11 | C | |
| 4 | AN | 78.8% | 70.0% | 11 | C | |
| 5 | DE | 76.7% | 60.0% | 11 | C | |
| 6 | AN | 73.8% | 50.0% | 11 | C | |
| 7 | AN | 73.1% | 40.0% | 11 | C | |
| 8 | BY | 70.2% | 30.0% | 11 | C | |
| 9 | ZA | 68.7% | 20.0% | 11 | C | |
| 10 | BY | 67.0% | 10.0% | 11 | C | |
| 11 | MA | 41.3% | 0.0% | 11 | C |
Score distribution
Top published rows on the benchmark's original scale.
CyberGym highlights
The top published results on this benchmark's own scale.
What is CyberGym?
Definition and scoring fields from the benchmark registry.
CyberGym is a benchmark for evaluating AI agents on cybersecurity tasks, testing their ability to identify vulnerabilities, perform security analysis, and complete security-related challenges in a controlled environment.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- CyberGym
- Modality
- text
- Primary category
- safety
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- cybergym|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about CyberGym.
Which model scores highest on CyberGym?
Gemini 3.5 Flash Cyber is currently ranked first with 83.2%.
What does CyberGym measure?
CyberGym is a benchmark for evaluating AI agents on cybersecurity tasks, testing their ability to identify vulnerabilities, perform security analysis, and complete security-related challenges in a controlled environment.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
11 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.