Claude Opus 4.8 — Real Scores Across 36 Benchmarks (2026)

Anthropic

Claude Opus 4.8

36 benchmarks

HMMT Feb 2026

#2 of 5

96.7%

HMMT Nov 2025

#1 of 4

96.5%

AIME

#19 of 117

95.7%

GPQA Diamond

#10 of 118

93.6%

GPQA Diamond

#10 of 118

91.0%

OmniDocBench

#27 of 31

87.9%

MATH-Vision

#5 of 15

86.7%

Terminal-Bench 2.1

#7 of 27

85.0%

Terminal-Bench 2.1

#7 of 27

84.6%

BrowseComp

#15 of 45

84.3%

MCP-Atlas

#5 of 17

83.6%

BabyVision

#5 of 5

81.2%

CharXiv

#13 of 23

80.5%

MMMU Pro

#14 of 36

78.9%

MCP-Atlas

#5 of 17

77.8%

Toolathlon-Verified

#2 of 7

76.2%

FrontierSWE

#3 of 8

75.1%

Program Bench

#4 of 6

71.9%

SWE-bench Pro

#4 of 43

69.2%

FrontierSWE

#3 of 8

66.7%

OfficeQA Pro

#2 of 6

63.9%

Toolathlon

#2 of 7

59.9%

DeepSWE

#11 of 22

59.0%

Humanity's Last Exam

#6 of 72

57.9%

OSWorld 2.0

#2 of 10

54.8%

Humanity's Last Exam

#6 of 72

49.8%

JobBench

#3 of 6

48.4%

PerceptionBench

#5 of 5

47.2%

MLS Bench

#4 of 6

42.8%

SWE Marathon

#2 of 6

40.0%

APEX-Agents

#4 of 7

39.4%

WorldVQA

#4 of 5

39.1%

PostTrain Bench

#5 of 6

34.1%

SpreadsheetBench 2

#4 of 6

31.55%

AutomationBench

#4 of 6

27.2%

ZeroBench

#5 of 5

17.0%