Claude Fable 5 — Real Scores Across 30 Benchmarks (2026)

Anthropic

Claude Fable 5

30 benchmarks

MATH-Vision

#2 of 15

94.8%

GPQA Diamond

#17 of 118

92.6%

BabyVision

#1 of 5

90.5%

OmniDocBench

#19 of 31

89.8%

CharXiv

#2 of 23

88.9%

BrowseComp

#8 of 45

88.0%

Terminal-Bench 2.1

#4 of 27

88.0%

FrontierSWE

#1 of 8

86.6%

OSWorld-Verified

#1 of 6

85.0%

MCP-Atlas

#1 of 17

84.7%

Terminal-Bench 2.1

#4 of 27

84.6%

MMMU Pro

#7 of 36

81.2%

SWE-bench Pro

#1 of 43

80.3%

Toolathlon-Verified

#1 of 7

77.9%

Program Bench

#3 of 6

76.8%

DeepSWE

#3 of 22

70.0%

OfficeQA Pro

#1 of 6

69.9%

Humanity's Last Exam

#2 of 72

59.0%

JobBench

#1 of 6

57.4%

PerceptionBench

#3 of 5

57.2%

WorldVQA

#1 of 5

56.7%

Humanity's Last Exam

#2 of 72

53.3%

MLS Bench

#1 of 6

49.9%

APEX-Agents

#1 of 7

43.3%

PostTrain Bench

#1 of 6

41.4%

SWE Marathon

#4 of 6

35.0%

SpreadsheetBench 2

#2 of 6

34.7%

FrontierCode Diamond

#1 of 1

29.3%

AutomationBench

#3 of 6

29.1%

ZeroBench

#2 of 5

23.0%