Benchmark Almanac

Kimi K2.5

Kimi K2.5 ranks first on 4 of the 227 benchmarks with five or more models scored.

Moonshot AI Released January 27th, 2026 Family Kimi K2 The lab's model card Compare with others
233
Benchmarks with a score
4
Ranked first, out of five or more models
19
In the top three
Moonshot AI Jan 2026

Kimi K2.5

Open weights Flagship

148.8 #75 of 256
Epoch Capabilities Index
Self-Reported Index 149.6 #99 of 404
  • Humanity's Last Exam 24.4% #17 /47
  • SWE-bench Verified 73.8% #18 /32
  • MMMU-Pro 78.5% #23 /90

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

BenchmarkScoreRankSource
MMLU-Pro 87.1% #9 of 167 Lab model card
Fiction.LiveBench 86.1% #7 of 57 Fiction.live leaderboard, via Epoch AI
Seal-0 57.4% #1 of 7 Lab model card
LMArena (Elo) 1,450 #64 of 335 Official leaderboard
EQ-Bench 3 1,530 #16 of 78 Official leaderboard
AA-LCR (Artificial Analysis Long Context Reasoning) 70% #8 of 38 Lab model card
KINA 40.2% #8 of 37 Official leaderboard
Longform Creative Writing (EQ-Bench) 74.9 #28 of 119 Official leaderboard
BullshitBench 47.3% #29 of 107 Official leaderboard
LongBench v2 61% #11 of 35 Lab model card
HUMAINE 30 #16 of 50 Official leaderboard
SimpleVQA 71.2% #10 of 30 Lab model card
MultiChallenge 61.4% #15 of 42 Official leaderboard
Humanity's Last Exam 24.4% #17 of 47 Official leaderboard
BullshitBench V2 52% #48 of 129 Official leaderboard
CL-bench 19.3% #10 of 24 Official leaderboard
Creative Writing v3 70.4 #52 of 120 Official leaderboard
MultiNRC 35.2% #18 of 38 Official leaderboard
Phare 67% #30 of 62 Official leaderboard
AdvancedIF 75.6% #3 of 6 Lab model card
WideSearch 72.7% #10 of 19 Lab model card
DesignArena ASCII Art 1,160 #41 of 77 Official leaderboard
CL-bench Life 13.2% #9 of 16 Official leaderboard
Extended NYT Connections 69.9% #58 of 103 Official leaderboard
SkateBench v2 57.2% #13 of 21 Official leaderboard
Antidote 1,006 #23 of 36 Official leaderboard
SimpleQA Verified 34.3% #62 of 94 Epoch AI run
Visual Aesthetic Benchmark 26.8% #12 of 18 Official leaderboard
Hemingway-bench 988 #30 of 42 Official leaderboard
Lech Mazur Creative Writing -1.13 #36 of 50 Official leaderboard
Document Arena 1,438 #33 of 39 Official leaderboard
Vectara Hallucination Leaderboard (2025 dataset) 14.2% #91 of 104 Official leaderboard
LOCA-Bench 4.00% #8 of 9 Official leaderboard

Math and science

BenchmarkScoreRankSource
ChemBench 67.2% #3 of 24 Official leaderboard
AIME 2025 96.1% #20 of 143 Lab model card
BRUMO 2025 98.3% #7 of 42 Official leaderboard
CMIMC 2025 90.6% #7 of 33 Official leaderboard
HMMT 2025 95.4% #12 of 54 Lab model card
SMT 2025 90.6% #10 of 40 Official leaderboard
OTIS Mock AIME 2024-2025 92.2% #62 of 212 Epoch AI run
FrontierMath 27.9% #21 of 70 Epoch AI run
GPQA Diamond 87.6% #69 of 229 Epoch AI run
AIME 2026 95.8% #13 of 31 Official leaderboard
General365 56.4% #10 of 23 Official leaderboard
FrontierMath Tier 4 4.20% #26 of 55 Epoch AI run
MathArena (Apex, IMO, USAMO) 8.85% #23 of 47 Official leaderboard
IMO-AnswerBench 81.8% #21 of 37 Lab model card
ARC-AGI-2 11.8% #58 of 98 Official leaderboard
SimpleBench 46.8% #63 of 106 Official leaderboard
ARC-AGI 65.3% #57 of 95 Official leaderboard
HMMT 2026 87.1% #19 of 31 Official leaderboard
MathArena Apex Shortlist 57.5% #25 of 38 Official leaderboard
HMMT Nov 2025 89.2% #15 of 22 Official leaderboard
ResearchClawBench 14.0 #19 of 23 Official leaderboard
Echo General AI Prediction Leaderboard 920 #17 of 17 Official leaderboard
KellyBench -89.60% #5 of 5 Official leaderboard

Coding

BenchmarkScoreRankSource
LiveSQLBench 32.2% #7 of 36 Official leaderboard
PaperBench 63.5% #3 of 15 Lab model card
StructEval 77.3% #6 of 28 Official leaderboard
LiveCodeBench v6 85% #20 of 90 Lab model card
SciCode 48.7% #11 of 49 Lab model card
Spider 2.0 55.9% #7 of 23 Official leaderboard
ts-bench v1 96% #3 of 9 Official leaderboard
AgenticDataBench 48.8% #1 of 3 Official leaderboard
OJBench 57% #3 of 8 Official leaderboard
DesignArena Website 1,242 #61 of 151 Official leaderboard
DesignArena Code 1,227 #62 of 143 Official leaderboard
DesignArena UI Component 1,228 #62 of 139 Official leaderboard
IOI (Vals) 17.7% #25 of 55 Official leaderboard
DesignArena Agentic Image to HTML 1,211 #5 of 11 Official leaderboard
DesignArena Game Dev 1,197 #68 of 143 Official leaderboard
DesignArena 3D Design 1,188 #65 of 136 Official leaderboard
DAXBench v2 79.7% #23 of 47 Official leaderboard
DesignArena Data Visualization 1,217 #58 of 118 Official leaderboard
SWE-bench Multilingual 73% #37 of 73 Lab model card
WeirdML 45.6% #61 of 112 Official leaderboard
OpenHands Index 49.2% #18 of 33 Official leaderboard
WebDev Arena 1,436 #58 of 105 arena.ai, via Epoch AI
Vision2Web S2-2026 33.2% #5 of 9 Official leaderboard
SWE-bench Verified 73.8% #18 of 32 Epoch AI run
Code Arena 1,436 #74 of 128 Official leaderboard
DesignArena Video to Website 1,179 #11 of 19 Official leaderboard
DesignArena SVG 1,140 #60 of 103 Official leaderboard
ALE-Bench 1,279 #75 of 126 Official leaderboard
DesignArena Agentic Overall 1,109 #48 of 74 Official leaderboard
DesignArena React Native 1,141 #37 of 56 Official leaderboard
FrontierCode 6.86% #8 of 12 Official leaderboard
DesignArena Full Stack 1,092 #34 of 49 Official leaderboard
DesignArena Frontend Web App 1,117 #39 of 53 Official leaderboard
PostTrainBench 10.3% #17 of 23 Official leaderboard
DesignArena Android (Kotlin) 1,105 #35 of 46 Official leaderboard
Vibe Code Bench 17.5% #93 of 120 Official leaderboard
TasteVal 8.60× #16 of 20 Official leaderboard
DesignArena Image to HTML 1,154 #42 of 51 Official leaderboard
DesignArena Image to Frontend 1,112 #26 of 31 Official leaderboard
SWE-Bench Pro 50.7% #68 of 81 Lab model card
APEX-SWE 30.3% #55 of 65 Official leaderboard
Image-to-WebDev Arena 1,439 #47 of 55 Official leaderboard
Code Migration 6.96% #67 of 78 Official leaderboard
FrogsGame post-training task 0% #15 of 17 Official leaderboard
FrontierSWE v1 23.4% #15 of 17 Official leaderboard
SWE Atlas Refactoring 20.9% #15 of 17 Official leaderboard
SlopCodeBench 9.69% #20 of 22 Official leaderboard
SWE Atlas 13.1% #20 of 22 Official leaderboard
RoadmapBench 11.3% #10 of 11 Official leaderboard
GameDevBench 20.7% #16 of 17 Official leaderboard
SWE Atlas Test Writing 25.8% #21 of 22 Official leaderboard
Vals RSI Index 0.07 #27 of 27 Official leaderboard

Agents and tools

BenchmarkScoreRankSource
OSWorld-MCP 56.6% #1 of 9 Official leaderboard
PinchBench 75.2% #11 of 48 Official leaderboard
OSWorld 63.3% #2 of 8 OS World Website, via Epoch AI
OpenThoughts-TBLite 75.1% #1 of 4 Official leaderboard
LisanBench 1851.7 #44 of 133 Official leaderboard
OSWorld-Verified 63.3% #13 of 33 Official leaderboard
SnakeBench 25 #89 of 188 Official leaderboard
Claw-Eval-Live 53.3% #6 of 12 Official leaderboard
LLM Chess 111 #62 of 123 Official leaderboard
Pencil Puzzle Bench 3.30% #27 of 53 Official leaderboard
Terminal-Bench 2.0 43.2% #22 of 43 Official leaderboard
BrowseComp 74.9% #42 of 81 Lab model card
AutoResearchBench 4.69% #7 of 13 Official leaderboard
ClawProBench 58.5% #27 of 49 Official leaderboard
MobileWorld 40.1% #10 of 18 Official leaderboard
Claw-Eval 74.9% #15 of 24 Official leaderboard
WebArena 58.9% #5 of 8 Lab model card
Clocktower Radio 1,517 #15 of 23 Official leaderboard
Toolathlon 27.8% #33 of 50 Official leaderboard
ScienceBoard 7.76% #10 of 15 Official leaderboard
GameWorld 37.4% #15 of 22 Official leaderboard
MCP Atlas 64.4% #28 of 37 Official leaderboard
DeepSearchQA 77.1% #18 of 23 Lab model card
PinchBench 2.0 54.6% #45 of 57 Official leaderboard
Claw-Eval Pass^3 52.3% #19 of 24 Official leaderboard
BalatroBench 5.33 #13 of 16 Official leaderboard
QwenClawBench 51.9% #10 of 12 Official leaderboard
CocoaBench 11.5% #5 of 6 Official leaderboard
RuneBench 2.10 #70 of 76 Official leaderboard
DeepPlanning 14.3% #14 of 15 Official leaderboard
WildClawBench 30.8% #29 of 31 Official leaderboard
Toolathlon-Verified 33% #25 of 25 Official leaderboard
Gaia2-CLI 34% #7 of 7 Official leaderboard
ClawBench 17.6% #6 of 6 Official leaderboard
CUA-World-Long 5.50% #4 of 4 Official leaderboard
Swarm Bench 58.3% #2 of 2 Lab model card
xbench-DeepSearch-2510 46% #2 of 2 Official leaderboard
xbench-DeepSearch 46% #1 of 1 Official leaderboard

Professional work

BenchmarkScoreRankSource
CorpFin v2 68.3% #10 of 119 Official leaderboard
Vals Finance Agent v1 49.2% #7 of 47 Official leaderboard
TaxEval v2 74.2% #39 of 129 Official leaderboard
ProfBench 51.8% #24 of 72 Official leaderboard
MortgageTax 66.5% #31 of 87 Official leaderboard
FinSearchComp 67.8% #3 of 7 Lab model card
CaseLaw v2 58.7% #28 of 53 Official leaderboard
Vals Finance Agent 50.6% #28 of 49 Official leaderboard
prinzbench 35.0 #15 of 26 Official leaderboard
StrataBench Land v1 Extract 87.9% #3 of 5 Official leaderboard
PRBench 43.8% #30 of 48 Official leaderboard
OccuBench 64.1% #9 of 14 Official leaderboard
Vals Index v1 51.7% #22 of 34 Official leaderboard
MedCode 39.3% #64 of 96 Official leaderboard
MedScribe 76.4% #69 of 97 Official leaderboard
SpreadsheetBench 2 14.6% #6 of 8 Official leaderboard
Legal Research Bench 15.9% #72 of 93 Official leaderboard
Vending-Bench 2 $1,198 #52 of 67 Official leaderboard
Workspace-Bench-Lite 49.1% #7 of 9 Official leaderboard
YC-Bench $408,822 #46 of 59 Official leaderboard
AlphaEval 43.9% #4 of 5 Official leaderboard
Harvey LAB (Legal Agent Benchmark) 0% #65 of 80 Official leaderboard
Vals Index v2 26.3% #56 of 65 Official leaderboard
Vals Finance Agent v2 35.8% #69 of 80 Official leaderboard
Excel Modeling Benchmark 28.5% #66 of 76 Official leaderboard
FORTE 55.6% #19 of 21 Official leaderboard
FoodTruck Bench $30 #56 of 59 Official leaderboard
EnterpriseBench: CoreCraft 8.70% #49 of 51 Official leaderboard

Vision, audio and robotics

BenchmarkScoreRankSource
InfographicVQA 92.6% #1 of 26 Lab model card
MathVista 90.1% #3 of 70 Lab model card
MMMU 84.3% #5 of 103 Lab model card
OCRBench 92.3% #3 of 34 Lab model card
BLINK 78.9% #3 of 29 Lab model card
VideoMME v2 (with subtitles) 40.7% #4 of 31 Official leaderboard
MMVU 80.4% #4 of 26 Lab model card
Video-MMMU 86.6% #6 of 36 Lab model card
CC-OCR-Bench-V2 73.3% #3 of 15 Official leaderboard
MMVP 87% #1 of 5 Lab model card
Video-MME 87.4% #12 of 51 Lab model card
MMMU-Pro 78.5% #23 of 90 Lab model card
LongVideoBench 79.8% #3 of 11 Lab model card
Vision Arena 1,252 #42 of 136 Official leaderboard
SpatialWorld Physical 9.20% #3 of 9 Official leaderboard
VisFactor 27.3% #9 of 26 Official leaderboard
MATH-Vision (MathVision) 84.2% #18 of 51 Lab model card
OmniDocBench v1.5 89.3 #4 of 11 Official leaderboard
LVBench 75.9% #12 of 31 Lab model card
MDPBench 77.5 #5 of 12 Official leaderboard
SpatialWorld Digital 31% #4 of 9 Official leaderboard
Math Kangaroo 2025 (MathArena) 80.6% #11 of 22 Official leaderboard
CiteVQA 19.1% #8 of 16 Official leaderboard
Claw-Eval-MM 50.2% #6 of 11 Official leaderboard
MotionBench 70.4% #5 of 9 Lab model card
CharXiv 77.5% #43 of 75 Lab model card
VISTA 41.9% #31 of 54 Official leaderboard
BabyVision 36.5% #16 of 27 Official leaderboard
WorldBench 52.5% #5 of 8 Official leaderboard
MineBench 1,187 #58 of 75 Official leaderboard
BVB 59.6 #37 of 45 Official leaderboard
Chartography 12.6% #36 of 43 Official leaderboard
ZeroBench 9.00% #27 of 31 Lab model card
OmniDocBench v1.6 84.5 #7 of 8 Official leaderboard
WorldVQA 46.3% #9 of 9 Lab model card

Safety and security

BenchmarkScoreRankSource
Agentic BAIM-LLM Evaluation (ABLE) 82.5% #3 of 18 Official leaderboard
LiveSecBench 74.8 #6 of 34 Official leaderboard
nocot-bench (NCRI) 80.6 #59 of 185 Official leaderboard
BioTIER Permit 99.4% #34 of 69 Official leaderboard
BinaryAudit 24.6% #10 of 20 Official leaderboard
EvalDetectBench 0.81 #5 of 10 Official leaderboard
KORA 38.5% #23 of 42 Official leaderboard
Human Pathogen Capabilities Test 49.2% #49 of 86 Official leaderboard
SMDD-Bench 30.3% #4 of 7 Official leaderboard
CyberGym 41.3% #21 of 34 Official leaderboard
ABC Bench Liquid-Handling Robot 73.3% #22 of 34 Official leaderboard
FORTRESS 41.1% #51 of 73 Official leaderboard
World-Class Bio 37.5% #52 of 74 Official leaderboard
Human Pathogen Capabilities Test v2 (HPCT v2) 43.4% #53 of 71 Official leaderboard
Molecular Biology Capabilities Test 42.4% #69 of 90 Official leaderboard
BioTIER Refuse 21.2% #57 of 69 Official leaderboard
SEC-bench Pro 2.20% #5 of 6 Official leaderboard
SEC-bench Pro 260617 2.30% #5 of 6 Official leaderboard
ABC Bench Fragment Design 74.8% #23 of 27 Official leaderboard
ABC Bench Screening Evasion 43.5% #15 of 17 Official leaderboard

Other

BenchmarkScoreRankSource
SAGE (Vals) 49.9% #19 of 81 Official leaderboard
TutorBench 54.6% #6 of 20 Official leaderboard
CAIS Text Capabilities Index 24.9 #45 of 68 Official leaderboard
CAIS Risk Index 65.5 #59 of 67 Official leaderboard

Its lab's numbers put it at 149.6, in line with its ECI

The Self-Reported Index fits one score to the 40 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 148.0 to 151.2. How we compute it.

Its lab's numbers next to 15 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
Humanity's Last Exam 50.2% Best of HLE-Full / HLE-Full (w/ tools) 24.4% Leaderboard +26 pts
Terminal-Bench 2.0 50.8% Terminal Bench 2.0 43.2% Leaderboard, Terminus 2 +7.6 pts
SWE-bench Verified 76.8% SWE-Bench Verified 70.8% Leaderboard, high, mini-SWE-agent +6.0 pts
SWE-bench Multilingual 73% SWE-Bench Multilingual 67.3% Leaderboard, mini-SWE-agent +5.7 pts
SimpleQA Verified 36.9% SimpleQA Verified (Table 4) 34.3% Epoch AI run +2.6 pts
HMMT 2025 95.4% HMMT 2025 (Feb) 93.3% Leaderboard, high +2.1 pts
HMMT Nov 2025 91.1% HMMT Nov 2025, average 24k output tokens (Table 5) 89.2% Leaderboard, high +1.9 pts
MATH-Vision (MathVision) 84.2% MathVision 85% Leaderboard, thinking, with python −0.8 pts
OJBench 57.4% OJBench (cpp) 57% Leaderboard, thinking +0.4 pts
AIME 2025 96.1% AIME 2025 95.8% Leaderboard, high +0.3 pts
BabyVision 36.5% BabyVision (Table 4) 36.5% Leaderboard Same
CyberGym 41.3% CyberGym 41.3% Leaderboard, Kimi Agent Same
GPQA Diamond 87.6% GPQA-Diamond 87.6% Epoch AI run, Fireworks Same
OSWorld-Verified 63.3% OSWorld-Verified, GUI actions only (Table 4) 63.3% Leaderboard, 100 steps Same
WorldVQA 46.3% WorldVQA 46.3% 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.

50 100 150 2023 2024 2025 2026 2027 Kimi K3.1 Pro Kimi K2.7 Code Kimi K2 Thinking 50 100 150 2023 2024 2025 2026 2027 Kimi K3.1 Pro Kimi K2.5

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 K2.5 model card (Hugging Face) Jan 27, 2026

Its system card

Every result in the longer documents the lab published with it. 11 of them aren't in the launch post.

Kimi K2.5: Visual Agentic Intelligence (arXiv 2602.02276) Jan 30, 2026

Capabilities

Hallucination

Cyber risk

AI R&D risk