Claude Opus 4.6 — Real Scores Across 18 Benchmarks (2026)

Anthropic

Claude Opus 4.6

18 benchmarks

AIME

#2 of 39

100.0%

MRCR v2

#2 of 5

93.0%

GPQA Diamond

#8 of 49

91.3%

BigLaw Bench

#1 of 1

90.2%

ARC-AGI 2

#1 of 6

85.0%

BrowseComp

#4 of 17

84.0%

MMLU Pro

#17 of 29

82.0%

Terminal-Bench 2.0

#3 of 14

81.8%

SWE-bench Verified

#2 of 40

81.42%

SWE-bench Multilingual

#2 of 3

77.8%

LiveCodeBench

#16 of 28

76.0%

OSWorld

#5 of 16

72.7%

Aider Polyglot

#6 of 7

72.0%

CyberGym

#3 of 3

66.6%

Humanity's Last Exam

#5 of 24

53.1%

USAMO 2026

#3 of 3

42.3%

SWE-bench Multimodal

#2 of 2

27.1%

ARC-AGI 3

#3 of 4

0.25%