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 — November 19, 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%, as reported by Xiaomi, followed by MiMo-V2.6-Pro (94.0%, as reported by Xiaomi) and DeepSeek V4.1 Flash 1017 (88.1%, as reported by DeepSeek). We do not use these results to rank models overall.
MiMo-V2.6-Flash
Xiaomi
Reported by XiaomiMiMo-V2.6-Pro
Xiaomi
Reported by XiaomiDeepSeek V4.1 Flash 1017
DeepSeek
Reported by DeepSeek29 modelsAgenticCurrentDisplay onlyUpdated November 19, 2026
Benchmark score table (29 models)
Of 29 results: 22 lab-reported, 7 benchmark-run.
Ranks on this table count the 7 benchmark-run and independent results. A lab-reported result keeps its place by score, without a rank.
- Lab-reported
- 22
- Independent
- 0
- Benchmark-run
- 7
Showing all 29 results, in score order.
| Rank | Model / configuration | Score | Who ran it | Parameters (B) | Open / closed |
|---|---|---|---|---|---|
| No rank on this table | MiMo-V2.6-FlashXiaomiReported by Xiaomi | 95.1% | Reported by Xiaomi | Not reported | Open |
| No rank on this table | MiMo-V2.6-ProXiaomiReported by Xiaomi | 94.0% | Reported by Xiaomi | Not reported | Open |
| No rank on this table | DeepSeek V4.1 Flash 1017DeepSeekReported by DeepSeek | 88.1% | Reported by DeepSeek | Not reported | Open |
| No rank on this table | Ling 3.1 FlashInclusionAIReported by InclusionAI | 87.9% | Reported by InclusionAI | Not reported | Open |
| No rank on this table | Fugu CyberSakana AIReported by Sakana AI | 86.9% | Reported by Sakana AI | Not reported | Closed |
| No rank on this table | Atria Dawn PreviewShanghai Artificial Intelligence LaboratoryReported by the lab | 86.5% | Reported by the lab | Not reported | Open |
| No rank on this table | Gemini 3.8 Flash CyberGoogleReported by Google | 86.2% | Reported by Google | Not reported | Closed |
| No rank on this table | Step 5 PreviewStepFunReported by StepFun | 84.7% | Reported by StepFun | Not reported | Open |
| No rank on this table | GPT-5.6 SolOpenAIReported by OpenAI | 84.5% | Reported by OpenAI | Not reported | Closed |
| No rank on this table | GLM-5.3Z.AIReported by Z.AI | 84.5% | Reported by Z.AI | Not reported | Open |
| No rank on this table | Claude Mythos 5AnthropicReported by Anthropic | 83.8% | Reported by Anthropic | Not reported | Closed |
| No rank on this table | DeepSeek V4 ProDeepSeekReported by DeepSeek | 83.3% | Reported by DeepSeek | Not reported | Open |
| No rank on this table | Gemini 3.5 Flash CyberGoogleReported by Google | 83.2% | Reported by Google | Not reported | Closed |
| 1 | Claude Mythos PreviewAnthropicRun by CyberGym | 83.1% | Run by CyberGym | Not reported | Closed |
| No rank on this table | MiniMax M3.1 FlashMiniMaxReported by MiniMax | 82.2% | Reported by MiniMax | Not reported | Closed |
| No rank on this table | GPT-5.6 TerraOpenAIReported by OpenAI | 81.8% | Reported by OpenAI | Not reported | Closed |
| No rank on this table | GPT-5.5OpenAIReported by OpenAI | 81.8% | Reported by OpenAI | Not reported | Closed |
| No rank on this table | GPT-5.4OpenAIReported by OpenAI | 79.0% | Reported by OpenAI | Not reported | Closed |
| No rank on this table | Hy4 previewTencentReported by Tencent | 78.4% | Reported by Tencent | Not reported | Open |
| No rank on this table | GPT-5.6 LunaOpenAIReported by OpenAI | 77.9% | Reported by OpenAI | Not reported | Closed |
| No rank on this table | DeepSeek V4 Flash 0731DeepSeekReported by DeepSeek | 76.7% | Reported by DeepSeek | Not reported | Open |
| No rank on this table | Claude Opus 4.7AnthropicReported by OpenAI (comparison table) | 73.1% | Reported by OpenAI (comparison table) | Not reported | Closed |
| 2 | GLM-5.1Z.AIRun by CyberGym | 68.7% | Run by CyberGym | Not reported | Open |
| 3 | Claude Opus 4.6AnthropicRun by CyberGym | 66.6% | Run by CyberGym | Not reported | Closed |
| 4 | Claude Sonnet 4.6AnthropicRun by CyberGym | 65.2% | Run by CyberGym | Not reported | Closed |
| No rank on this table | Muse Spark 1.1MetaReported by Meta | 59.0% | Reported by Meta | Not reported | Closed |
| 5 | Claude Opus 4.5AnthropicRun by CyberGym | 50.6% | Run by CyberGym | Not reported | Closed |
| 6 | Muse SparkMetaRun by CyberGym | 43.5% | Run by CyberGym | Not reported | Closed |
| 7 | GLM-5Z.AIRun by CyberGym | 43.2% | Run by CyberGym | Not reported | Open |
A dash in the rank column marks a lab-reported result. It keeps its place by score, without a rank on this table.
- Lab-reported
- Benchmark-run
- Independent
Results reported by model developersUnranked
Labs ran these themselves, with a metric, split or setting the table above does not use. They take no rank here.
| Model | Reported value | What was measured | Against the table above | Who ran it |
|---|---|---|---|---|
| MAI-Cyber-1-FlashMicrosoft | 90.4%Target any-of | CyberGym Level 1MDASH harness with GPT-5.4 as the escalation model, a two-model system | Not comparable to the scores above | Reported by Microsoft |
A lab’s own result is shown for reference. It never takes a rank on this board.
How we label sourcesAmong the reported CyberGym rows, MiMo-V2.6-Flash is first at 95.1%, as reported by Xiaomi. 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.
CyberGym is one of the cyber benchmarks we track. The cybersecurity benchmark comparison sets it beside the others, one table per security job.
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, as reported by Xiaomi.
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 7,000+ subscribers.
One email each week. Unsubscribe anytime.