GPT-5.5 vs Claude Opus 4.8 — Who Actually Wins? Real Benchmark Data (2026)

head to head

GPT-5.5 vs Claude Opus 4.8

33 shared benchmarks

13

wins

0

ties

20

wins

38.5%

APEX-Agents

39.4%

0.43%

ARC-AGI 3

1.5%

46.9%

Agents' Last Exam

45.2%

22.7%

AutomationBench

27.2%

83.6%

BabyVision

81.2%

84.4%

BrowseComp

84.3%

84.1%

CharXiv

80.5%

67.0%

DeepSWE

59.0%

47.9%

ExploitBench

40.0%

51.7%

FrontierMath

80.0%

35.4%

FrontierMath Tier 4

56.1%

64.9%

FrontierSWE

66.7%

93.5%

GPQA Diamond

91.0%

41.4%

Humanity's Last Exam

49.8%

38.3%

JobBench

48.4%

92.2%

MATH-Vision

86.7%

82.8%

MCP-Atlas

77.8%

35.5%

MLS Bench

42.8%

81.2%

MMMU Pro

78.9%

49.5%

OSWorld 2.0

54.8%

60.9%

OfficeQA Pro

63.9%

89.4%

OmniDocBench

87.9%

55.8%

PerceptionBench

47.2%

28.4%

PostTrain Bench

34.1%

70.8%

Program Bench

71.9%

14.0%

SWE Marathon

40.0%

58.6%

SWE-bench Pro

69.2%

29.05%

SpreadsheetBench 2

31.55%

83.4%

Terminal-Bench 2.1

84.6%

55.6%

Toolathlon

59.9%

73.5%

Toolathlon-Verified

76.2%

38.5%

WorldVQA

39.1%

22.0%

ZeroBench

17.0%