OpenAI · model
GPT-6 Cyber benchmark scores
As of 2026-11-05, GPT-6 Cyber (OpenAI) has measured scores on 4 benchmarks and estimated scores on 14 more.
| Benchmark | Score | Source |
|---|---|---|
| GDPval-AA | 56.7% | estimated ± 7.4 pp, low confidence |
| BrowseComp | 95.2% | estimated ± 3.0 pp, low confidence |
| CyberGym | 86.6% | estimated ± 10.0 pp, low confidence |
| AA Agentic Index | 57.5% | estimated ± 8.0 pp, low confidence |
| AA-AnalystAgent | 52.3% | estimated ± 4.6 pp, medium confidence |
| MazeBench | 22.4% | estimated ± 3.0 pp, medium confidence |
| AA AutomationBench | 74.4% | estimated ± 6.2 pp, medium confidence |
| DeepsecBench-AA | 37.3% | estimated ± 8.1 pp, low confidence |
| GDP.pdf | 30.9% | estimated ± 7.2 pp, low confidence |
| AA ITBench | 53.9% | estimated ± 4.3 pp, medium confidence |
| CWE-bench v1 | 70.1% | estimated ± 4.5 pp, medium confidence |
| Hermes Bench | 73.8% | estimated ± 0.3 pp, low confidence |
| Hermes Index | 59.6% | estimated ± 2.6 pp, low confidence |
| ExploitBench | 100.0% | measured |
| ExploitGym | 48.5% | measured |
| Agents' Last Exam | 51.7% | estimated ± 9.6 pp, low confidence |
| SEC-Bench Pro | 87.0% | measured |
| ExploitBench Internal Port | 42.0% | measured |