CyberGym
We show this table for reference; we do not rank on it.
A cybersecurity task benchmark for evaluating defensive cyber workflows and vulnerability-oriented agent performance.
Benchmark score on CyberGym — October 20, 2026
We compile the CyberGym rows from benchmark-owner or independent runs, provider self-reports and secondary reports. MiMo-V2.6-Flash leads the table at 95.1%, followed by MiMo-V2.6-Pro (94.0%) and DeepSeek V4.1 Flash 1017 (88.1%). We do not use these results to rank models overall.
MiMo-V2.6-Flash
Xiaomi
MiMo-V2.6-Pro
Xiaomi
DeepSeek V4.1 Flash 1017
DeepSeek
29 modelsAgenticCurrentDisplay onlyUpdated October 20, 2026
Benchmark score table (29 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | MiMo-V2.6-FlashXiaomi | 95.1% | Not reported | Open |
| 2 | MiMo-V2.6-ProXiaomi | 94.0% | Not reported | Open |
| 3 | DeepSeek V4.1 Flash 1017DeepSeek | 88.1% | Not reported | Open |
| 4 | Ling 3.1 FlashInclusionAI | 87.9% | Not reported | Open |
| 5 | Fugu CyberSakana AI | 86.9% | Not reported | Closed |
| 6 | Atria Dawn PreviewShanghai Artificial Intelligence Laboratory | 86.5% | Not reported | Open |
| 7 | Gemini 3.8 Flash CyberGoogle | 86.2% | Not reported | Closed |
| 8 | Step 5 PreviewStepFun | 84.7% | Not reported | Open |
| 9 | GPT-5.6 SolOpenAI | 84.5% | Not reported | Closed |
| 10 | GLM-5.3Z.AI | 84.5% | Not reported | Open |
| 11 | Claude Mythos 5Anthropic | 83.8% | Not reported | Closed |
| 12 | DeepSeek V4 Pro 0813DeepSeek | 83.3% | Not reported | Open |
| 13 | Gemini 3.5 Flash CyberGoogle | 83.2% | Not reported | Closed |
| 14 | Claude Mythos PreviewAnthropic | 83.1% | Not reported | Closed |
| 15 | MiniMax M3.1 FlashMiniMax | 82.2% | Not reported | Closed |
| 16 | GPT-5.6 TerraOpenAI | 81.8% | Not reported | Closed |
| 17 | GPT-5.5OpenAI | 81.8% | Not reported | Closed |
| 18 | GPT-5.4OpenAI | 79.0% | Not reported | Closed |
| 19 | Hy4 previewTencent | 78.4% | Not reported | Open |
| 20 | GPT-5.6 LunaOpenAI | 77.9% | Not reported | Closed |
| 21 | DeepSeek V4 Flash 0731DeepSeek | 76.7% | Not reported | Open |
| 22 | Claude Opus 4.7 (Adaptive)Anthropic | 73.1% | Not reported | Closed |
| 23 | GLM-5.1Z.AI | 68.7% | Not reported | Open |
| 24 | Claude Opus 4.6Anthropic | 66.6% | Not reported | Closed |
| 25 | Claude Sonnet 4.6Anthropic | 65.2% | Not reported | Closed |
| 26 | Muse Spark 1.1Meta | 59.0% | Not reported | Closed |
| 27 | Claude Opus 4.5Anthropic | 50.6% | Not reported | Closed |
| 28 | Muse SparkMeta | 43.5% | Not reported | Closed |
| 29 | GLM-5Z.AI | 43.2% | Not reported | Open |
Among the reported CyberGym rows, MiMo-V2.6-Flash is first at 95.1%. The third row is 7.0 points behind. The broader top-10 range is 10.6 points, so the table still separates the published systems.
29 models have been evaluated on CyberGym. The benchmark falls in the Agentic category. CyberGym is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About CyberGym
Year
2026
Tasks
1,507 vulnerability analysis instances
Format
Vulnerability reproduction and PoC generation
Difficulty
Real-world cybersecurity
CyberGym includes 1,507 benchmark instances from historical vulnerabilities across 188 large software projects. BenchLM stores CyberGym as a display-only agentic security benchmark when exact provider comparison values are published.
Freshness and provenance
Version
CyberGym 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does CyberGym measure?
A cybersecurity task benchmark for evaluating defensive cyber workflows and vulnerability-oriented agent performance.
Which model scores highest on CyberGym?
MiMo-V2.6-Flash by Xiaomi currently leads with a score of 95.1% on CyberGym.
How many models are evaluated on CyberGym?
29 AI models have published results on CyberGym in the BenchLM catalog.
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 5,600+ readers.
One email each week. Unsubscribe anytime.