benchgap
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.

BenchmarkScoreSource
GDPval-AA57.8%estimated ± 7.6 pp, low confidence
BrowseComp92.3%estimated ± 3.0 pp, medium confidence
CyberGym86.5%estimated ± 10.0 pp, low confidence
AA Agentic Index52.5%estimated ± 8.0 pp, low confidence
Chess Puzzles68.8%estimated ± 8.0 pp, medium confidence
AA-AnalystAgent50.9%estimated ± 5.5 pp, low confidence
Furniture Assembly77.4%estimated ± 6.5 pp, medium confidence
MazeBench16.4%estimated ± 3.8 pp, high confidence
Mystery Game Puzzles77.5%estimated ± 7.5 pp, medium confidence
NYT Connections94.3%estimated ± 10.3 pp, low confidence
Terminal-Bench 4.164.6%estimated ± 5.0 pp, medium confidence
CWE-bench v165.4%estimated ± 4.5 pp, medium confidence
FrontierPhysics15.4%estimated ± 2.3 pp, high confidence
StealthMark65.5%estimated ± 6.1 pp, medium confidence
Terminal-Bench 4.070.5%measured
Terminal-Bench-Science 0.175.3%measured
AA AutomationBench75.0%estimated ± 4.8 pp, medium confidence
GDP.pdf27.3%estimated ± 7.4 pp, low confidence
AA Terminal-Bench 4.065.9%estimated ± 6.3 pp, medium confidence
CWE-bench64.1%estimated ± 9.0 pp, low confidence
MKA50.6%estimated ± 3.1 pp, high confidence
FACTS Multimodal51.2%estimated ± 2.7 pp, medium confidence
Hermes Bench74.8%estimated ± 1.5 pp, medium confidence
Hermes Index58.6%estimated ± 5.5 pp, low confidence
ExploitBench95.0%measured
ExploitGym38.1%measured
DAYJOB11.9%estimated ± 1.1 pp, medium confidence
DAYJOB Healthcare2.7%estimated ± 1.5 pp, medium confidence
Agents' Last Exam51.7%estimated ± 9.6 pp, medium confidence
SEC-Bench Pro83.3%estimated ± 3.9 pp, high confidence
ExploitBench Internal Port31.6%estimated ± 12.3 pp, low confidence