GPT-5.4 vs Claude Opus 4.6 — Who Actually Wins? Real Benchmark Data (2026)

head to head

GPT-5.4 vs Claude Opus 4.6

14 shared benchmarks

7

wins

0

ties

7

wins

73.3%

ARC-AGI 2

85.0%

0.26%

ARC-AGI 3

0.5%

91.0%

BigLaw Bench

90.2%

82.7%

BrowseComp

84.0%

79.0%

CyberGym

66.6%

92.8%

GPQA Diamond

91.3%

21.4%

GraphWalks BFS

38.7%

36.24%

Humanity's Last Exam

40.0%

36.6%

MRCR v2

93.0%

75.0%

OSWorld

72.7%

57.7%

SWE-bench Pro

53.4%

77.2%

SWE-bench Verified

81.42%

75.1%

Terminal-Bench 2.0

65.4%

95.2%

USAMO 2026

42.3%