Claude Opus 4.8 vs GPT-5.6 Sol — Who Actually Wins? Real Benchmark Data (2026)

head to head

Claude Opus 4.8 vs GPT-5.6 Sol

32 shared benchmarks

8

wins

1

ties

23

wins

39.4%

APEX-Agents

39.9%

1.5%

ARC-AGI 3

7.78%

45.2%

Agents' Last Exam

52.7%

27.2%

AutomationBench

29.7%

81.2%

BabyVision

88.9%

84.3%

BrowseComp

90.4%

80.5%

CharXiv

84.6%

59.0%

DeepSWE

72.7%

40.0%

ExploitBench

73.5%

80.0%

FrontierMath

83.0%

66.7%

FrontierSWE

71.3%

91.0%

GPQA Diamond

94.6%

49.8%

Humanity's Last Exam

44.5%

48.4%

JobBench

46.5%

86.7%

MATH-Vision

95.8%

77.8%

MCP-Atlas

83.6%

42.8%

MLS Bench

46.2%

78.9%

MMMU Pro

83.0%

54.8%

OSWorld 2.0

62.6%

63.9%

OfficeQA Pro

63.2%

87.9%

OmniDocBench

85.8%

47.2%

PerceptionBench

59.7%

34.1%

PostTrain Bench

34.6%

71.9%

Program Bench

77.6%

40.0%

SWE Marathon

39.0%

69.2%

SWE-bench Pro

64.6%

31.55%

SpreadsheetBench 2

32.4%

84.6%

Terminal-Bench 2.1

88.8%

59.9%

Toolathlon

58.0%

76.2%

Toolathlon-Verified

74.9%

39.1%

WorldVQA

41.8%

17.0%

ZeroBench

17.0%