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 11, 2026
Models11
Model coverage11
MetricScore
EvidenceB
CyberGym Ranking
Higher score ranks better on this benchmark.
| 01 | GO | 83.2% | 100.0% | 11 | C | |
| 02 | AN | 83.1% | 90.0% | 11 | C | |
| 03 | OP | 81.8% | 80.0% | 11 | C | |
| 04 | AN | 78.8% | 70.0% | 11 | C | |
| 05 | DE | 76.7% | 60.0% | 11 | C | |
| 06 | AN | 73.8% | 50.0% | 11 | C | |
| 07 | AN | 73.1% | 40.0% | 11 | C | |
| 08 | BY | 70.2% | 30.0% | 11 | C | |
| 09 | ZA | 68.7% | 20.0% | 11 | C | |
| 10 | BY | 67.0% | 10.0% | 11 | C | |
| 11 | MA | 41.3% | 0.0% | 11 | C |
CyberGym Score Distribution
A closer view of the leading scores on this benchmark.
CyberGym Highlights
The leading models and scores on this benchmark.
What is CyberGym?
What CyberGym measures and how its scores work.
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. This benchmark is not independently verified and has an evidence level of B.
- Family
- CyberGym
- Modality
- text
- Primary category
- safety
- Score direction
- higher
- LLMBoard eligible
- Yes
- Evaluation key
- cybergym|llm-stats-current
Benchmark scores retain their original unit. Overall score eligibility is shown separately.
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 model results are currently shown.
Does this benchmark affect the overall score?
Yes. This benchmark can contribute to the current LLMBoard capability score.