Benchmark
CyberGym
CyberGym is a benchmark for evaluating AI agents on cybersecurity tasks, testing their ability to identify vulnerabilities, perform security analysis, and…
- Modality
- text
- Categories
- safety, agents, code
- Openness
- unknown
- Source
- llm_stats
- Reported scores
- 13
Reported scores
llm_stats
| Model | Organization | Reported value | Reported | Evidence |
|---|---|---|---|---|
| Claude Mythos Preview | Anthropic | 0.831 | 2026-04-07 | evidence |
| Claude Opus 4.6 | Anthropic | 0.738 | 2026-02-05 | evidence |
| Claude Opus 4.7 | Anthropic | 0.731 | 2026-04-16 | evidence |
| Claude Opus 4.8 | Anthropic | 0.788 | 2026-05-28 | evidence |
| DeepSeek-V4-Flash-0731 | DeepSeek | 0.767 | 2026-07-31 | evidence |
| DeepSeek-V4-Pro-0813 | DeepSeek | 0.833 | 2026-08-13 | evidence |
| Gemini 3.5 Flash Cyber | 0.832 | 2026-07-21 | evidence | |
| GLM-5.1 | Z.ai | 0.687 | 2026-04-07 | evidence |
| GLM-5.3 | Z.ai | 0.845 | 2026-08-14 | evidence |
| GPT-5.5 | OpenAI | 0.818 | 2026-04-23 | evidence |
| Kimi K2.5 | Moonshot AI | 0.413 | 2026-01-27 | evidence |
| Seed 2.1 Pro | ByteDance | 0.702 | 2026-06-24 | evidence |
| Seed 2.1 Turbo | ByteDance | 0.67 | 2026-06-24 | evidence |
Scores are partitioned by the source that reported them and are never merged into a single cross-source ranking, because the sources measure different things and say so.