Kimi K2 Jul 2025

Kimi K2 ranks first on 3 of the 78 benchmarks with five or more models scored.

Moonshot AI Released July 12th, 2025 Family Kimi K2 The lab's model card Compare with others

Benchmarks with a score

Ranked first, out of five or more models

In the top three

Moonshot AI Jul 2025

Kimi K2

Jul 2025 snapshot Open weights Flagship

141.2 #116 of 256

Epoch Capabilities Index

Self-Reported Index 132.7 #253 of 404

SWE-Bench Pro 27.7% #17 /24
LongBench v2 49.1% #25 /35
SuperGPQA 57.2% #23 /35

Where it ranks

Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.

Knowledge and language

Benchmark	Score	Rank	Source
DROP	93.5%	#1 of 41	Lab model card
Vectara Hallucination Leaderboard (HHEM)	1.10%	#5 of 148	Official leaderboard
WritingBench	81.3	#3 of 45	Official leaderboard
FACTS Grounding	88.5%	#1 of 15	Lab model card
EQ-Bench 3	1,548	#11 of 78	Official leaderboard
MMLU	89.5%	#19 of 131	Lab model card
IFEval	89.8%	#31 of 135	Lab model card
Creative Writing v3	75.0	#30 of 120	Official leaderboard
LMArena (Elo)	1,418	#107 of 335	Official leaderboard
SimpleQA	31%	#22 of 60	Lab model card
Longform Creative Writing (EQ-Bench)	68.1	#48 of 119	Official leaderboard
MMLU-Redux	92.7%	#19 of 45	Lab model card
MMLU-Pro	81.1%	#71 of 167	Lab model card
OpenAI MRCR	55%	#5 of 10	Lab model card
Judgemark v2.1	73.1	#22 of 42	Official leaderboard
Fiction.LiveBench	61.1%	#34 of 57	Fiction.live leaderboard, via Epoch AI
SuperGPQA	57.2%	#23 of 35	Lab model card
MultiChallenge	54.1%	#28 of 42	Lab model card
WideSearch	3.00%	#4 of 6	Official leaderboard
LLM Confabulation Leaderboard	20.4%	#39 of 58	Official leaderboard
LongBench v2	49.1%	#25 of 35	Lab model card
MultiNRC	18.5%	#29 of 38	Official leaderboard
FRAMES	77.1%	#7 of 9	Lab model card
Arena-Hard v2.0	54.5%	#19 of 24	Lab model card
Humanity's Last Exam	4.70%	#153 of 154	Lab model card
FaithJudge	7.40%	#1 of 1	Lab model card

Math and science

Benchmark	Score	Rank	Source
AutoLogi	89.5%	#1 of 8	Lab model card
MATH-500	97.4%	#14 of 70	Lab model card
GPQA Main	65.3%	#18 of 60	Official leaderboard
PolyMath	65.1%	#8 of 24	Lab model card
ZebraLogic	89%	#8 of 23	Lab model card
GPQA Diamond	75.1%	#149 of 302	Lab model card
HiPhO	54.4%	#20 of 36	Official leaderboard
PHYBench	21.3%	#16 of 28	Official leaderboard
ForecastBench	0.14	#56 of 82	Official leaderboard
AIME 2024	69.6%	#67 of 88	Lab model card
SimpleBench	26.3%	#90 of 106	Official leaderboard
CNMO 2024	74.3%	#6 of 7	Lab model card
AIME 2025	49.5%	#132 of 143	Lab model card
HMMT 2025	38.8%	#53 of 54	Lab model card

Coding

Benchmark	Score	Rank	Source
MultiPL-E	85.7%	#2 of 18	Lab model card
AutoCodeBench	47.8%	#10 of 34	Official leaderboard
SWE-bench Verified	65.4%	#22 of 60	Official leaderboard
Aider Polyglot	60%	#15 of 30	Lab model card
PaperBench	27.8%	#9 of 15	Lab model card
ArtifactsBench	47.6%	#13 of 20	Official leaderboard
SWE-Bench Pro	27.7%	#17 of 24	Official leaderboard
WeirdML	39.4%	#81 of 112	Official leaderboard
GSO-Bench	4.90%	#22 of 30	Official leaderboard
Multi-SWE-bench flash	18.3%	#3 of 4	Official leaderboard
LiveCodeBench v6	53.7%	#78 of 90	Lab model card
OJBench	27.1%	#14 of 16	Lab model card
Multi-SWE-bench	18.3%	#9 of 10	Lab model card
SWE-Lancer	39.1%	#9 of 10	Lab model card
SWE-bench Multilingual	47.3%	#68 of 73	Lab model card
IOI (Vals)	1.25%	#52 of 55	Official leaderboard

Agents and tools

Benchmark	Score	Rank	Source
AceBench	82.2%	#2 of 16	Official leaderboard
BFCL V4	59.1%	#9 of 65	Official leaderboard
Galileo Agent Leaderboard v2	53%	#5 of 21	Official leaderboard
Terminal-Bench	27.8%	#9 of 18	tbench.ai, via Epoch AI
LisanBench	347.3	#77 of 133	Official leaderboard
MCPMark	19.1%	#17 of 28	Official leaderboard
Elimination Game	3	#40 of 56	Official leaderboard
tau2-bench	66.1%	#63 of 79	Lab model card
Step Race	1	#49 of 59	Official leaderboard
MCP-Universe	19.1%	#27 of 30	Official leaderboard
LLM Chess	-689	#119 of 123	Official leaderboard
lmgame-Bench	6.28	#23 of 23	Official leaderboard

Professional work

Benchmark	Score	Rank	Source
LegalBench	81.5%	#78 of 138	Official leaderboard
TaxEval v2	70.2%	#86 of 129	Official leaderboard
CorpFin v2	50.4%	#97 of 119	Official leaderboard
PRBench	36.4%	#45 of 48	Official leaderboard

Safety and security

Benchmark	Score	Rank	Source
HELM Safety	98.0	#7 of 75	Official leaderboard
SimpleSafetyTests	100%	#16 of 75	Official leaderboard
AIR-Bench 2024	74.1%	#30 of 82	Official leaderboard
Anthropic Red Team	99.3%	#30 of 75	Official leaderboard
LiveSecBench	35.6	#23 of 34	Official leaderboard
FORTRESS	55.5%	#64 of 73	Official leaderboard

Other

Benchmark	Score		Rank	Source
HELM Capabilities	76.8		#10 of 60	Official leaderboard
LiveBench	76.4%		#7 of 27	Lab model card

Its lab's numbers put it at 132.7, 8.5 points below its ECI

The Self-Reported Index fits one score to the 33 capability results Moonshot AI reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 130.8 to 134.6. How we compute it.

Its lab's numbers next to 10 leaderboards

Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.

Benchmark	Lab says	Board says	Gap
AceBench	76.5% AceBench, Acc	82.2% Leaderboard	−5.7 pts
IFEval	89.8% IFEval, Prompt Strict	85% Leaderboard, strict, share of instructions followed	+4.8 pts
LongBench v2	49.1% LongBench v2 (Acc.), 128K window (Table 3)	44.9% Leaderboard	+4.2 pts
MATH-500	97.4% MATH-500, Acc	94.2% Leaderboard, Vals AI harness	+3.2 pts
tau2-bench	66.1% τ2-Bench micro Pass@1 across domains (Appendix C text). Avg@4: retail 70.6, airline 56.5, telecom 65.8 (Table 3)	64.3% Leaderboard	+1.8 pts
Aider Polyglot	60% Aider-Polyglot, Acc	59.1% Leaderboard, aider, diff edits	+0.9 pts
SuperGPQA	57.2% SuperGPQA, Acc	58.1% Leaderboard	−0.9 pts
SWE-bench Verified	65.8% SWE-bench Verified (Agentic Coding), Single Attempt (Acc)	65.4% Leaderboard, OpenHands	+0.4 pts
MMLU-Pro	81.1% MMLU-Pro, EM	81% Leaderboard	Same
Vectara Hallucination Leaderboard (HHEM)	1.10% HHEM v2.1 (1-Hallu.), hallucination rate 1.1% (Table 3)	1.10% Leaderboard	Same

Moonshot AI's models over time

Moonshot AI's models in color, against every other model on the Epoch Capabilities Index, by release date.

Source: Epoch Capabilities Index, Epoch AI (CC BY 4.0)

Its launch post

The benchmarks the lab chose to headline, with the scores it gave.

Kimi K2 Instruct model card (Hugging Face) Jul 11, 2025

Its system card

Every result in the longer documents the lab published with it. 14 of them aren't in the launch post.

Kimi K2: Open Agentic Intelligence (arXiv 2507.20534) Jul 21, 2025