Anthropic · model
Claude Mythos 5.2 benchmark scores
As of 2026-11-18, Claude Mythos 5.2 (Anthropic) has measured scores on 4 benchmarks and estimated scores on 27 more.
| Benchmark | Score | Source |
|---|---|---|
| GDPval-AA | 57.8% | estimated ± 7.6 pp, low confidence |
| BrowseComp | 92.3% | estimated ± 3.0 pp, medium confidence |
| CyberGym | 86.5% | estimated ± 10.0 pp, low confidence |
| AA Agentic Index | 52.5% | estimated ± 8.0 pp, low confidence |
| Chess Puzzles | 68.8% | estimated ± 8.0 pp, medium confidence |
| AA-AnalystAgent | 50.9% | estimated ± 5.5 pp, low confidence |
| Furniture Assembly | 77.4% | estimated ± 6.5 pp, medium confidence |
| MazeBench | 16.4% | estimated ± 3.8 pp, high confidence |
| Mystery Game Puzzles | 77.5% | estimated ± 7.5 pp, medium confidence |
| NYT Connections | 94.3% | estimated ± 10.3 pp, low confidence |
| Terminal-Bench 4.1 | 64.6% | estimated ± 5.0 pp, medium confidence |
| CWE-bench v1 | 65.4% | estimated ± 4.5 pp, medium confidence |
| FrontierPhysics | 15.4% | estimated ± 2.3 pp, high confidence |
| StealthMark | 65.5% | estimated ± 6.1 pp, medium confidence |
| Terminal-Bench 4.0 | 70.5% | measured |
| Terminal-Bench-Science 0.1 | 75.3% | measured |
| AA AutomationBench | 75.0% | estimated ± 4.8 pp, medium confidence |
| GDP.pdf | 27.3% | estimated ± 7.4 pp, low confidence |
| AA Terminal-Bench 4.0 | 65.9% | estimated ± 6.3 pp, medium confidence |
| CWE-bench | 64.1% | estimated ± 9.0 pp, low confidence |
| MKA | 50.6% | estimated ± 3.1 pp, high confidence |
| FACTS Multimodal | 51.2% | estimated ± 2.7 pp, medium confidence |
| Hermes Bench | 74.8% | estimated ± 1.5 pp, medium confidence |
| Hermes Index | 58.6% | estimated ± 5.5 pp, low confidence |
| ExploitBench | 95.0% | measured |
| ExploitGym | 38.1% | measured |
| DAYJOB | 11.9% | estimated ± 1.1 pp, medium confidence |
| DAYJOB Healthcare | 2.7% | estimated ± 1.5 pp, medium confidence |
| Agents' Last Exam | 51.7% | estimated ± 9.6 pp, medium confidence |
| SEC-Bench Pro | 83.3% | estimated ± 3.9 pp, high confidence |
| ExploitBench Internal Port | 31.6% | estimated ± 12.3 pp, low confidence |