Claude Mythos Preview — Real Scores Across 16 Benchmarks (2026)

Anthropic

Claude Mythos Preview

16 benchmarks

Cybench

#1 of 1

100.0%

USAMO 2026

#1 of 4

97.6%

GPQA Diamond

#3 of 129

94.6%

SWE-bench Verified

#1 of 88

93.9%

SWE-bench Multilingual

#1 of 16

87.3%

BrowseComp

#10 of 45

86.9%

CyberGym

#1 of 6

83.1%

Terminal-Bench 2.0

#2 of 30

82.0%

GraphWalks BFS

#1 of 3

80.0%

OSWorld

#1 of 32

79.6%

SWE-bench Pro

#3 of 50

77.8%

ExploitBench

#3 of 8

74.2%

DeepSWE

#3 of 23

69.7%

Humanity's Last Exam

#1 of 77

64.7%

Toolathlon

#3 of 13

61.1%

SWE-bench Multimodal

#1 of 3

59.0%