GPT-5.6 Sol — Real Scores Across 35 Benchmarks (2026)

OpenAI

GPT-5.6 Sol

35 benchmarks

MATH-Vision

#1 of 15

95.8%

GPQA Diamond

#2 of 118

94.6%

GPQA Diamond

#2 of 118

94.1%

BrowseComp

#3 of 45

90.4%

FrontierMath

#1 of 11

89.0%

BabyVision

#2 of 5

88.9%

Terminal-Bench 2.1

#2 of 27

88.8%

OmniDocBench

#29 of 31

85.8%

CharXiv

#6 of 23

84.6%

MCP-Atlas

#4 of 17

83.6%

FrontierMath

#1 of 11

83.0%

MMMU Pro

#3 of 36

83.0%

Program Bench

#2 of 6

77.6%

Toolathlon-Verified

#3 of 7

74.9%

ExploitBench

#2 of 2

73.5%

DeepSWE

#1 of 22

73.0%

DeepSWE

#1 of 22

72.7%

FrontierSWE

#5 of 8

71.3%

SEC-Bench Pro

#2 of 2

71.2%

SWE-bench Pro

#6 of 43

64.6%

OfficeQA Pro

#4 of 6

63.2%

OSWorld 2.0

#1 of 10

62.6%

PerceptionBench

#1 of 5

59.7%

JobBench

#4 of 6

46.5%

MLS Bench

#3 of 6

46.2%

Humanity's Last Exam

#28 of 72

44.5%

WorldVQA

#3 of 5

41.8%

APEX-Agents

#3 of 7

39.9%

SWE Marathon

#3 of 6

39.0%

PostTrain Bench

#3 of 6

34.6%

ExploitGym

#1 of 1

33.7%

SpreadsheetBench 2

#3 of 6

32.4%

AutomationBench

#2 of 6

29.7%

ZeroBench

#4 of 5

17.0%

ARC-AGI 3

#1 of 8

7.78%