GPT-5.5 vs GPT-5.6 Sol — Who Actually Wins? Real Benchmark Data (2026)

head to head

GPT-5.5 vs GPT-5.6 Sol

35 shared benchmarks

2

wins

0

ties

33

wins

38.5%

APEX-Agents

39.9%

0.43%

ARC-AGI 3

7.78%

46.9%

Agents' Last Exam

52.7%

22.7%

AutomationBench

29.7%

83.6%

BabyVision

88.9%

84.4%

BrowseComp

90.4%

88.1%

Capture-the-Flag Challenges

96.7%

84.1%

CharXiv

84.6%

67.0%

DeepSWE

72.7%

47.9%

ExploitBench

73.5%

15.1%

ExploitGym

33.7%

51.7%

FrontierMath

83.0%

64.9%

FrontierSWE

71.3%

93.5%

GPQA Diamond

94.6%

41.4%

Humanity's Last Exam

44.5%

38.3%

JobBench

46.5%

92.2%

MATH-Vision

95.8%

82.8%

MCP-Atlas

83.6%

35.5%

MLS Bench

46.2%

81.2%

MMMU Pro

83.0%

49.5%

OSWorld 2.0

62.6%

60.9%

OfficeQA Pro

63.2%

89.4%

OmniDocBench

85.8%

55.8%

PerceptionBench

59.7%

28.4%

PostTrain Bench

34.6%

70.8%

Program Bench

77.6%

45.8%

SEC-Bench Pro

71.2%

14.0%

SWE Marathon

39.0%

58.6%

SWE-bench Pro

64.6%

29.05%

SpreadsheetBench 2

32.4%

83.4%

Terminal-Bench 2.1

88.8%

55.6%

Toolathlon

58.0%

73.5%

Toolathlon-Verified

74.9%

38.5%

WorldVQA

41.8%

22.0%

ZeroBench

17.0%