Kimi K2.6

Kimi K2.6 ranks first on EvoArena PersonaMem-Evo.

Moonshot AI Released April 20th, 2026 Family Kimi K2 The lab's model card Compare with others

207

Benchmarks with a score

Ranked first, out of five or more models

In the top three

Moonshot AI Apr 2026

Kimi K2.6

Open weights Flagship

151.6 #62 of 256

Epoch Capabilities Index

Self-Reported Index 153.6 #69 of 404

SWE-bench Verified 76.7% #10 /32
MMMU-Pro 79.4% #19 /90
AutomationBench 2.59% #38 /42

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
Phare	76.1%	#7 of 62	Official leaderboard
EvoArena PersonaMem-Evo	55.5%	#1 of 8	Official leaderboard
EQ-Bench 3	1,548	#10 of 78	Official leaderboard
LMArena (Elo)	1,461	#49 of 335	Official leaderboard
Longform Creative Writing (EQ-Bench)	78.5	#18 of 119	Official leaderboard
Humanity's Last Exam	54%	#29 of 154	Lab model card
WideSearch	80.8%	#4 of 19	Lab model card
Wer wird Millionär? LLM Benchmark	114156.0	#17 of 77	Official leaderboard
BullshitBench V2	65%	#29 of 129	Official leaderboard
Creative Writing v3	74.7	#33 of 120	Official leaderboard
CompassArena	1,047	#3 of 10	Official leaderboard
Extended NYT Connections	87.2%	#31 of 103	Official leaderboard
Antidote	1,045	#12 of 36	Official leaderboard
HUMAINE	29	#20 of 50	Official leaderboard
Hemingway-bench	1,021	#21 of 42	Official leaderboard
EvoArena SWE-Chain-Evo	30.2%	#4 of 8	Official leaderboard
EvoArena Terminal-Bench-Evo	42.9%	#4 of 8	Official leaderboard
Lech Mazur Creative Writing	0.03	#26 of 50	Official leaderboard
Judgemark v4	57.3	#22 of 40	Official leaderboard
DesignArena ASCII Art	1,152	#44 of 77	Official leaderboard
StudyArena Text	1,225	#50 of 85	Official leaderboard
SimpleQA Verified	34.9%	#61 of 94	Epoch AI run
EQ-Bench 4	1,191	#21 of 32	Official leaderboard
Vectara Hallucination Leaderboard (2025 dataset)	10.8%	#69 of 104	Official leaderboard
Document Arena	1,450	#28 of 39	Official leaderboard

Math and science

Benchmark	Score	Rank	Source
Echo General AI Prediction Leaderboard	1,017	#2 of 17	Official leaderboard
FrontierMath	39%	#8 of 55	Official leaderboard
GPQA Diamond	90.8%	#41 of 229	Epoch AI run
OTIS Mock AIME 2024-2025	96.1%	#43 of 212	Epoch AI run
DTBench	90.9%	#42 of 180	Official leaderboard
FrontierMath Tier 4	14.6%	#15 of 55	Epoch AI run
MathArena (Apex, IMO, USAMO)	24%	#13 of 47	Official leaderboard
MathArena Apex Shortlist	77.1%	#11 of 38	Official leaderboard
HMMT 2026	94.7%	#9 of 31	Official leaderboard
IMO-AnswerBench	86%	#11 of 37	Lab model card
AIME 2026	95.8%	#16 of 31	Official leaderboard
FrontierMath Tiers 1-3 v2	57.2%	#45 of 86	Official leaderboard
USAMO 2026	51.2%	#5 of 9	Official leaderboard
ProofBench	16%	#35 of 59	Official leaderboard
ResearchClawBench	18.0	#14 of 23	Official leaderboard
FrontierMath Tier 4 v2	25.6%	#42 of 68	Official leaderboard
NatureBench	7.78%	#12 of 18	Official leaderboard
xbench-ScienceQA	65.8%	#25 of 29	Official leaderboard
HorizonMath	0.90%	#8 of 8	Official leaderboard

Coding

Benchmark	Score	Rank	Source
LiveCodeBench v6	89.6%	#7 of 90	Lab model card
OJBench	60.6%	#2 of 16	Lab model card
SciCode	52.2%	#7 of 49	Lab model card
DesignArena Website	1,264	#39 of 151	Official leaderboard
LiveSQLBench	29.3%	#10 of 36	Official leaderboard
DesignArena 3D Design	1,250	#38 of 136	Official leaderboard
DesignArena Code	1,263	#42 of 143	Official leaderboard
AutoLab	0.46	#3 of 10	Official leaderboard
SWE-bench Verified	76.7%	#10 of 32	Epoch AI run
DesignArena UI Component	1,258	#47 of 139	Official leaderboard
SWE-bench Multilingual	76.7%	#25 of 73	Lab model card
WeirdML	55.9%	#39 of 112	Official leaderboard
ALE-Bench	1,695	#44 of 126	Official leaderboard
DesignArena Data Visualization	1,247	#42 of 118	Official leaderboard
DesignArena Game Dev	1,240	#51 of 143	Official leaderboard
DesignArena Agentic Image to HTML	1,222	#4 of 11	Official leaderboard
EvoCode-Bench	31.9%	#5 of 13	Official leaderboard
SWE-Bench Pro	58.6%	#32 of 81	Lab model card
WebDev Arena	1,509	#42 of 105	arena.ai, via Epoch AI
OpenHands Index	57.1%	#14 of 33	Official leaderboard
FrontierCS	46.9	#6 of 14	Official leaderboard
Code Arena	1,509	#58 of 128	Official leaderboard
EvoCode-Bench v2	13.2%	#6 of 13	Official leaderboard
DesignArena Video to Website	1,184	#9 of 19	Official leaderboard
DesignArena SVG	1,172	#50 of 103	Official leaderboard
FrontierCode	16%	#6 of 12	Official leaderboard
AndroidBench	67.6%	#18 of 32	Official leaderboard
Code Migration	27.8%	#44 of 78	Official leaderboard
DesignArena Agentic Overall	1,151	#43 of 74	Official leaderboard
Vibe Code Bench	37.9%	#74 of 120	Official leaderboard
Image-to-WebDev Arena	1,522	#34 of 55	Official leaderboard
DesignArena Full Stack	1,134	#31 of 49	Official leaderboard
SlopCodeBench	18.9%	#14 of 22	Official leaderboard
RoadmapBench	14.8%	#7 of 11	Official leaderboard
DesignArena React Native	1,149	#36 of 56	Official leaderboard
DesignArena Android (Kotlin)	1,151	#30 of 46	Official leaderboard
GameCraft-Bench	30.6%	#11 of 15	Official leaderboard
DesignArena Image to HTML	1,169	#39 of 51	Official leaderboard
FrogsGame post-training task	0.40%	#13 of 17	Official leaderboard
DesignArena Godot Game Dev	1,103	#19 of 24	Official leaderboard
GBAEval	0.86%	#17 of 21	Official leaderboard
Claw-SWE-Bench	66.9%	#13 of 16	Official leaderboard
DesignArena Agentic Game Dev	1,089	#22 of 27	Official leaderboard
MLS-Bench	26.7%	#14 of 15	Official leaderboard
FrontierSWE v1	22.2%	#16 of 17	Official leaderboard
SWE-Marathon	0%	#12 of 12	Official leaderboard
QwenSVG	1,325	#7 of 7	Official leaderboard
Vision2Web S3-2026	54.8%	#7 of 7	Official leaderboard

Agents and tools

Benchmark	Score	Rank	Source
Toolathlon	50%	#7 of 50	Official leaderboard
Clocktower Radio	1,696	#4 of 23	Official leaderboard
Terminal-Bench 2.0	66.7%	#12 of 63	Lab model card
DeepSearchQA	92.5%	#5 of 23	Lab model card
PACT	1,560	#6 of 26	Official leaderboard
Pencil Puzzle Bench	20%	#13 of 53	Official leaderboard
Claw-Eval	77.9%	#7 of 24	Official leaderboard
BrowseComp	83.2%	#26 of 81	Lab model card
OSWorld-Verified	73.1%	#11 of 33	Official leaderboard
MobileWorld	44.8%	#6 of 18	Official leaderboard
DeepTerminalBench	32%	#3 of 9	Official leaderboard
SkillsBench	54%	#8 of 22	Official leaderboard
UniClawBench	36.3%	#4 of 10	Official leaderboard
OpenComputer	58.8%	#3 of 7	Official leaderboard
GameHorizon	67.8	#17 of 37	Official leaderboard
Claw-Eval Pass^3	62.3%	#6 of 13	Lab model card
CivBench	1,504	#9 of 19	Official leaderboard
ClawProBench	59.3%	#24 of 49	Official leaderboard
MCPMark	55.9%	#5 of 10	Lab model card
Toolathlon-Verified	58%	#13 of 25	Official leaderboard
VitaBench 2.0	39.7%	#9 of 17	Official leaderboard
Harbor-Index	8.50%	#5 of 9	Official leaderboard
LiveClawBench	68.6%	#10 of 17	Official leaderboard
CEO-Bench	$43,598	#12 of 19	Official leaderboard
Mystery Game Puzzles	18%	#56 of 84	Epoch AI run
OSWorld 2.0	4.60%	#5 of 7	Official leaderboard
STATE-Bench Main 0.7.0	48.3%	#3 of 4	Official leaderboard
Chess Puzzles	20%	#72 of 93	Epoch AI run
TUA-Bench	42.8%	#10 of 12	Official leaderboard
RuneBench	3.60	#64 of 76	Official leaderboard
Long-Horizon-Terminal-Bench 2.0	42%	#13 of 15	Official leaderboard
MCPMark Verified	72.8%	#7 of 8	Official leaderboard
AutomationBench	2.59%	#38 of 42	Official leaderboard
TERMS-Bench	0.60	#10 of 11	Official leaderboard
Long-Horizon-Terminal-Bench (LHTB)	25.5%	#23 of 25	Official leaderboard
EBR-bench	2.38%	#31 of 31	Epoch AI run

Professional work

Benchmark	Score	Rank	Source
CorpFin v2	66.7%	#17 of 119	Official leaderboard
LegalBench	84.7%	#30 of 138	Official leaderboard
Vals Index v1	59.3%	#8 of 34	Official leaderboard
Vals Finance Agent	57.1%	#12 of 49	Official leaderboard
TaxEval v2	74.7%	#33 of 129	Official leaderboard
Vending-Bench 2	$6,205	#24 of 67	Official leaderboard
MortgageTax	65.8%	#36 of 87	Official leaderboard
CaseLaw v2	61.2%	#22 of 53	Official leaderboard
Excel Modeling Benchmark	57.9%	#39 of 76	Official leaderboard
LEXam-hard	36.2%	#8 of 15	Official leaderboard
DesignArena HTML Slides	1,080	#15 of 28	Official leaderboard
EnterpriseClawBench	56.1%	#5 of 9	Official leaderboard
Vals Index v2	43.5%	#40 of 65	Official leaderboard
PhysicianBench	17%	#8 of 13	Official leaderboard
Harvey LAB (Legal Agent Benchmark)	1.67%	#50 of 80	Official leaderboard
MedScribe	78.2%	#62 of 97	Official leaderboard
CHI-Bench	15.6%	#15 of 23	Official leaderboard
MedCode	40.1%	#63 of 96	Official leaderboard
FORTE	63.9%	#14 of 21	Official leaderboard
APEX-Agents	27.9%	#8 of 12	Lab model card
Vals Finance Agent v2	44.9%	#56 of 80	Official leaderboard
MerchantBench	25.0	#6 of 8	Official leaderboard
YC-Bench	$511,137	#45 of 59	Official leaderboard
Tax Agent Bench	12.4%	#65 of 84	Official leaderboard
Public Benefits Bench	56.6%	#39 of 50	Official leaderboard
Legal Research Bench	15.9%	#73 of 93	Official leaderboard
EnterpriseBench: CoreCraft	24.6%	#41 of 51	Official leaderboard
DesignArena Python-PPTX Slides	1,087	#30 of 36	Official leaderboard
Agents' Last Exam	9.21%	#26 of 31	Official leaderboard
ProfBench	41.4%	#61 of 72	Official leaderboard
E-Commerce Bench	0.70	#17 of 19	Official leaderboard
FrontierFinance	32.3%	#16 of 17	Official leaderboard
Tuesday Work Index	24.9	#43 of 44	Official leaderboard
PolicyBench	66.9%	#49 of 50	Official leaderboard
Kimi Claw 24/7 Bench	42.9%	#4 of 4	Official leaderboard

Vision, audio and robotics

Benchmark	Score	Rank	Source
GSM8K-V	47.3%	#2 of 24	Official leaderboard
V*	96.9%	#2 of 17	Lab model card
MMMU-Pro	79.4%	#19 of 90	Lab model card
Vision Arena	1,265	#33 of 136	Official leaderboard
MATH-Vision (MathVision)	87.4%	#14 of 51	Lab model card
CC-OCR-Bench-V2	72.9%	#5 of 15	Official leaderboard
BabyVision	68.5%	#10 of 27	Official leaderboard
CharXiv	80.4%	#35 of 75	Lab model card
MineBench	1,425	#44 of 75	Official leaderboard
QwenWorldBench	50.9%	#3 of 5	Official leaderboard
AgentWorldBench	53.4	#8 of 13	Official leaderboard
PerceptionBench (Moonshot)	42.6%	#11 of 16	Official leaderboard
Claw-Eval-MM	49.6%	#8 of 11	Official leaderboard
Chartography	12.2%	#37 of 43	Official leaderboard
Blueprint-Bench 2	3.90%	#33 of 38	Official leaderboard
Furniture Assembly Benchmark	21.7%	#33 of 35	Official leaderboard

Safety and security

Benchmark	Score	Rank	Source
Human Pathogen Capabilities Test	79%	#2 of 86	Official leaderboard
Molecular Biology Capabilities Test	59.8%	#8 of 90	Official leaderboard
World-Class Bio	54.1%	#21 of 74	Official leaderboard
nocot-bench (NCRI)	79.2	#61 of 185	Official leaderboard
CyberBench (Vals)	74.3%	#9 of 26	Official leaderboard
ABC Bench Screening Evasion	62.5%	#7 of 17	Official leaderboard
BioTIER Permit	99.6%	#29 of 69	Official leaderboard
PCT (Polarized Contrastive Pairs)	43.6%	#21 of 46	Official leaderboard
Human Pathogen Capabilities Test v2 (HPCT v2)	54.5%	#34 of 71	Official leaderboard
SecCodeBench	61.1%	#3 of 6	Official leaderboard
HarnessAudit-Bench	0.30	#5 of 7	Official leaderboard
BiomniBench	59.1	#6 of 8	Official leaderboard
ABC Bench Fragment Design	76.5%	#21 of 27	Official leaderboard
Agentic BAIM-LLM Evaluation (ABLE)	42.9%	#14 of 18	Official leaderboard
ExploitBench	18%	#7 of 9	Official leaderboard
BioTIER Refuse	21.9%	#54 of 69	Official leaderboard
ABC Bench Liquid-Handling Robot	55.3%	#27 of 34	Official leaderboard
VCT-v2	37.2%	#20 of 25	Official leaderboard

Other

Benchmark	Score	Rank	Source
SAGE (Vals)	50.2%	#17 of 81	Official leaderboard
ACCoRD	71.4	#45 of 152	Official leaderboard
Conceptual Reasoning Index	53.6	#57 of 151	Official leaderboard
CAIS Vision Capabilities Index	55.3	#22 of 53	Official leaderboard
LMCA	37.3%	#71 of 147	Official leaderboard
CAIS Text Capabilities Index	31.4	#35 of 68	Official leaderboard
Vals Multimodal Index	56.4%	#19 of 33	Official leaderboard
StudentBench	12.9	#7 of 11	Official leaderboard
CompassRank	75.7	#9 of 14	Official leaderboard
CAIS Risk Index	59.5	#51 of 67	Official leaderboard

Its lab's numbers put it at 153.6, 2.0 points above its ECI

The Self-Reported Index fits one score to the 25 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 151.6 to 155.6. How we compute it.

Its lab's numbers next to 9 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	What differs
BabyVision	39.8% BabyVision	68.5% Leaderboard, tools	−29 pts	The lab score excludes Python tools, while the board reports the Python-enabled run.
MATH-Vision (MathVision)	87.4% MathVision	93.2% Leaderboard, with python	−5.8 pts
SWE-bench Verified	80.2% SWE-Bench Verified	76.7% Epoch AI run	+3.5 pts
HMMT 2026	92.7% HMMT 2026 (Feb)	94.7% Leaderboard, high	−2.0 pts
AIME 2026	96.4% AIME 2026	95.8% Leaderboard, high	+0.6 pts
GPQA Diamond	90.5% GPQA-Diamond	90.8% Epoch AI run	−0.3 pts
OSWorld-Verified	73.1% OSWorld-Verified	73.1% Leaderboard, 100 steps	Same
Claw-Eval Pass^3	62.3% Claw Eval (pass^3)	62.3% Leaderboard	Same
Toolathlon	50% Toolathlon	50% 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 K2.6 model card (Hugging Face) Apr 20, 2026

Best of HLE-Full (w/ tools) / HLE-Full 54
BrowseComp 83.2
DeepSearchQA (f1-score) 92.5
WideSearch (item-f1) 80.8
Toolathlon 50
MCPMark 55.9
Claw Eval (pass^3) 62.3
APEX-Agents 27.9
OSWorld-Verified 73.1
Terminal-Bench 2.0 (Terminus-2) 66.7
SWE-Bench Pro 58.6
SWE-Bench Multilingual 76.7
SWE-Bench Verified 80.2
SciCode 52.2
OJBench (python) 60.6
LiveCodeBench (v6) 89.6
AIME 2026 96.4
HMMT 2026 (Feb) 92.7
IMO-AnswerBench 86
GPQA-Diamond 90.5
MMMU-Pro 79.4
CharXiv (RQ) 80.4
MathVision 87.4
BabyVision 39.8
V* (w/ python) 96.9