GLM-5.2 — Real Scores Across 27 Benchmarks (2026)

Z.ai

GLM-5.2

27 benchmarks

AIME

#4 of 117

99.2%

HMMT Nov 2025

#3 of 4

94.4%

HMMT Feb 2026

#4 of 5

92.5%

GPQA Diamond

#23 of 118

91.2%

Terminal-Bench 2.1

#13 of 27

82.7%

MCP-Atlas

#7 of 17

82.6%

Terminal-Bench 2.1

#13 of 27

81.0%

SWE-bench Verified

#11 of 86

80.0%

MCP-Atlas

#7 of 17

76.8%

FrontierSWE

#4 of 8

74.4%

FrontierSWE

#4 of 8

67.3%

Program Bench

#6 of 6

63.7%

SWE-bench Pro

#10 of 43

62.1%

Toolathlon-Verified

#6 of 7

59.9%

Humanity's Last Exam

#8 of 72

54.7%

Toolathlon

#7 of 7

48.2%

DeepSWE

#16 of 22

46.2%

DeepSWE

#16 of 22

44.0%

JobBench

#5 of 6

43.4%

OfficeQA Pro

#6 of 6

41.4%

Humanity's Last Exam

#8 of 72

40.5%

MLS Bench

#5 of 6

40.4%

APEX-Agents

#7 of 7

35.6%

PostTrain Bench

#4 of 6

34.3%

SpreadsheetBench 2

#6 of 6

28.12%

SWE Marathon

#6 of 6

13.0%

AutomationBench

#6 of 6

12.9%