Kimi K2.5 — Real Scores Across 14 Benchmarks (2026)

Moonshot AI

Kimi K2.5

14 benchmarks

HumanEval

#1 of 2

99.0%

AIME

#11 of 39

96.1%

GPQA Diamond

#18 of 49

87.6%

MMLU Pro

#6 of 29

87.1%

LiveCodeBench

#8 of 28

85.0%

MMMU Pro

#6 of 15

78.5%

BrowseComp

#7 of 17

78.4%

SWE-bench Verified

#16 of 40

76.8%

OSWorld

#9 of 16

63.3%

BrowseComp

#7 of 17

60.6%

WebArena

#3 of 3

58.9%

Terminal-Bench 2.0

#13 of 14

50.8%

SWE-bench Pro

#10 of 13

50.7%

Humanity's Last Exam

#19 of 24

24.37%