AIEvals

Models

Kimi K3

Moonshot AI’s Kimi K3 release, with independent coding, multimodal and agentic results.

Moonshot AIReleased 16 July 2026Context 1M tokensOutput 1M tokensOpen weights$3.00 in / $15.00 out per Mtok

Family Kimi K372 benchmarks with a published result1 independent first place among the 46 models this site listsRead 17 October 2026

Where it stands

Rank compares tracked models with a score. Published shows the source’s rank and field size, when available.

BenchmarkScoreRankPublishedLeaderBehind
Vals IndexVals AI50.70%25of 3326of 41Gemini 4 Argon18.20 pts
Artificial Analysis Intelligence IndexArtificial Analysis4416of 42no list readClaude Opus 5.514 pts
Terminal-Bench 4.0Artificial Analysis12.63%31of 40no list readClaude Sonnet 5.551.01 pts
SWE-bench (Verified)Vals AI93.40%8of 238of 88Claude Opus 53.60 pts
LiveCodeBenchVals AI87.19%12of 2118of 143Claude Fable 5.13.33 pts
GPQA DiamondVals AI92.93%11of 2312of 138Gemini 3.1 Pro2.52 pts
MMLU ProVals AI87.97%15of 2422of 138Claude Fable 5.14.41 pts
MMMU ProVals AI88.15%9of 2011of 93Claude Fable 5.12.49 pts
Vibe Code Bench v1.1Vals AI84.96%13of 3813of 106Claude Sonnet 5.57.43 pts
ProgramBenchVals AI2.00%6of 296of 54Claude Opus 5.516.50 pts
APEX-Agents-AAArtificial Analysis41.30%2of 7no list readGemini 3 5 Flash5.75 pts
τ³-BankingArtificial Analysis45.98%8of 30no list readQwen3.8 Max5.36 pts
ITBench-AAArtificial Analysis47.69%8of 18no list readGPT-5.6 Sol8.52 pts
AA-AnalystAgentArtificial Analysis38.75%14of 18no list readGemini 3.7 Flash21.25 pts
MLCR-AAArtificial Analysis38.33%11of 37no list readClaude Sonnet 5.536.67 pts
SWE-Bench Pro V2 FullScale AI / SEAL97.70%3of 10no list readClaude Opus 51.70 pts
SWE-Bench Pro V2 HARDScale AI / SEAL88.20%4of 10no list readClaude Opus 59.80 pts
Vals RSI IndexVals AI20.88%13of 2013of 23Claude Opus 5.516.43 pts
Vibe Code Bench 1–100Vals AI18.24%10of 2010of 20Claude Opus 5.512.12 pts
Harvey LAB-AAArtificial Analysis94.64%2of 20no list readMuse Spark 1.3 Max0.84 pts
EnterpriseOps-Gym-AAArtificial Analysis45.33%10of 22no list readClaude Fable 55.79 pts
AutomationBench-AAArtificial Analysis58.27%19of 40no list readGemini 4 Argon19.24 pts
SciCodeArtificial Analysis59.49%7of 40no list readClaude Opus 5.57.41 pts
GDP.pdfArtificial Analysis22.00%16of 40no list readGPT-6.1 Sol9.00 pts
CritPtArtificial Analysis23.43%15of 42no list readGPT-5.6 Sol8.86 pts
AA-OmniscienceArtificial Analysis19.722of 42no list readClaude Opus 5.526.72
AA-LCR v1.1Artificial Analysis88.67%1of 42no list readKimi K3—
Terminal-Bench 2.1Vals AI80.90%9of 359of 75Claude Opus 5.56.74 pts
ProofBench v1.1Vals AI87.00%9of 3410of 46Claude Sonnet 5.513.00 pts
LegalBenchVals AI86.02%11of 2715of 149Claude Fable 52.54 pts
Harvey's Legal Agent BenchmarkVals AI10.83%9of 3712of 73Muse Spark 1.214.59 pts
Vals Multimodal IndexVals AI73.42%3of 173of 33Claude Fable 50.73 pts
BioMysteryBenchVals AI71.48%9of 199of 21Claude Sonnet 5.59.63 pts
Code MigrationVals AI16.10%35of 3748of 71Claude Sonnet 5.553.73 pts
CyberBench v1.1Vals AI75.24%6of 316of 43GPT-6 Sol2.74 pts
EMBVals AI66.40%18of 3718of 69Claude Fable 5.110.27 pts
Finance Agent (v2)Vals AI54.36%18of 3723of 73Gemini 4 Argon11.04 pts
IOIVals AI48.94%23of 3124of 38Gemini 4 Argon51.06 pts
Legal Research BenchVals AI44.23%16of 3717of 72Muse Spark 1.3 Max11.06 pts
MedCodeVals AI48.88%14of 3525of 104Claude Opus 514.69 pts
MedScribeVals AI87.96%11of 3613of 106Claude Opus 5.53.47 pts
Public Benefits Bench v1.1Vals AI68.27%10of 2910of 45Claude Opus 58.66 pts
SAGEVals AI54.26%3of 324of 90Claude Opus 4.71.84 pts
Tax Agent BenchVals AI68.67%10of 3711of 64Claude Fable 5.18.97 pts
DeepSWE v1.1Datacurve68.51%8of 208of 28GPT-6 Astra5.61 pts
GDPval-AA v2.1Artificial Analysis1,539.9924of 42no list readClaude Opus 5.5306.18
AA-Briefcase v1.1Artificial Analysis1,500.516of 41no list readClaude Opus 5.5321.35
Humanity's Last ExamArtificial Analysis46.90%17of 42no list readClaude Opus 5.514.45 pts

Nearby on Vals Index

The two models above and below on the Vals Index (Vals AI), on every column where this model has a published number.

ModelVals Index%Artificial Analysis Intelligence IndexindexTerminal-Bench 4.0%SWE-bench%LiveCodeBench%GPQA Diamond%MMLU Pro%MMMU Pro%Vibe Code Bench v1.1%ProgramBench%
Gemini 3.7 FlashGoogle51.27%3913.64%80.80%88.65%93.94%90.12%88.96%70.39%0.00%
GPT-6 LunaOpenAI51.22%3712.63%–––––81.65%0.50%
Kimi K3Moonshot AI50.70%4412.63%93.40%87.19%92.93%87.97%88.15%84.96%2.00%
Hunyuan Hy4 PreviewTencent49.94%–––––––77.48%0.00%
Muse Spark 1.2Meta49.29%407.07%86.60%––88.28%86.13%79.10%–
Bold marks a figure better than this model’s on the same column, whichever model it belongs to. Ranks cover the models shown from the stated source.

What the hosted runtime reports

Ollama Cloud serves its own copy of this model, so its context cap and its prices are the ones a caller actually gets. Read 25 September 2026.

Parameters
2.81T parameters16 of 896 experts per token
Hosted context
1M tokens
Hosted price / Mtok
$3.00in · $0.300 cached · $15.00 out
Vision inputTool callingReasoning modeServed by the hostInput: Text, Image

Described by Moonshot as the first open 3T-class model, with weights released under the Kimi K3 licence.

ollama.com/library/kimi-k3

Results by benchmark

Independent and vendor results retain their own harnesses and effort settings.

BenchmarkIndependent runAs reported by a publisher
APEX-Agents-AAKnowledge work41.30%Artificial Analysisnot published
τ³-BankingTool use45.98%Artificial Analysis
37.1%tau3 bankingReflection AI · 5 October 2026Rank 3 of 6 in this publisher table
ITBench-AAIT operations47.69%Artificial Analysisnot published
AA-AnalystAgentData analysis38.75%Artificial Analysisnot published
MLCR-AAMedical reasoning38.33%Artificial Analysisnot published
SWE-Bench Pro V2 FullRepository coding97.70%Scale AI / SEALnot published
SWE-Bench Pro V2 HARDRepository coding88.20%Scale AI / SEAL
88.2%SWE Bench Pro v2-HardReflection AI · 5 October 2026Rank 1 of 4 in this publisher table
Vals RSI IndexAutonomous research20.88%Vals AInot published
Vibe Code Bench 1–100Application development18.24%Vals AInot published
Harvey LAB-AALegal reasoning94.64%Artificial Analysisnot published
EnterpriseOps-Gym-AAEnterprise workflows45.33%Artificial Analysisnot published
Apex Bench v1.1Reasoningno published result
47.8%Apex Bench v1.1DeepSeek · 17 October 2026Rank 6 of 6 in this publisher table
PLawBenchProfessional knowledgeno published result
76.2%PLawBenchZ.ai · 12 October 2026Rank 3 of 5 in this publisher table
SWE Atlas Codebase QnACodebase understandingno published result
68.0%SWE Atlas Codebase QnAReflection AI · 5 October 2026Rank 1 of 3 in this publisher table
AutomationBench (public split)Business workflowsno published result
46.7%AutomationBench publicReflection AI · 5 October 2026Rank 3 of 6 in this publisher table
MCP AtlasAgentic toolingno published result
82.3%MCP AtlasReflection AI · 5 October 2026Rank 3 of 7 in this publisher table
BrowseComp (context management)Web researchno published result
91.2%BrowseComp w/ context managementReflection AI · 5 October 2026Rank 1 of 4 in this publisher table
DeepSearchQA (context management)Web researchno published result
95.0%DeepSearchQA w/context managementReflection AI · 5 October 2026Rank 1 of 2 in this publisher table
AA-LCR (version not specified)Long-context reasoningno published result
88.7%AA-LCRReflection AI · 5 October 2026Rank 1 of 8 in this publisher table
AutomationBench-AAWorkflow automation58.27%Artificial Analysisnot published
SciCodeScientific coding59.49%Artificial Analysis
58.7%SciCodeReflection AI · 5 October 2026Rank 2 of 7 in this publisher table
GDP.pdfDocument reasoning22.00%Artificial Analysisnot published
CritPtScientific reasoning23.43%Artificial Analysis
23.4%CriPT AAReflection AI · 5 October 2026Rank 1 of 8 in this publisher table
AA-OmniscienceFactual knowledge19.7Artificial Analysisnot published
AA-LCR v1.1Long-context reasoning88.67%Artificial Analysisnot published
Terminal-Bench 4.0Agentic coding12.63%Artificial Analysis
19.70%Vals AI
12.6%Terminal-Bench 4.0DeepSeek · 17 October 2026Rank 6 of 6 in this publisher table
12.6%Terminal-Bench 4.0 (Pass@1)DeepSeek · 10 September 2026Rank 5 of 7 in this publisher table
Terminal-Bench 2.1Agentic coding80.90%Vals AI
85.02%Artificial Analysis
88.3%Terminal Bench v2.1Reflection AI · 5 October 2026Rank 2 of 8 in this publisher table
88.3%Terminal Bench 2.1Z.ai · 14 August 2026Rank 2 of 8 in this publisher table
88.3%Terminal-Bench 2.1 (Pass@1)DeepSeek · 10 September 2026Rank 4 of 7 in this publisher table
SWE-bench (Verified)Software engineering93.40%Vals AInot published
LiveCodeBenchCode generation87.19%Vals AInot published
GPQA DiamondScience reasoning92.93%Vals AI
93.54%Artificial Analysis
93.5%GPQA DiamondReflection AI · 5 October 2026Rank 1 of 8 in this publisher table
92.9%GPQA Diamond (Pass@1)DeepSeek · 10 September 2026Rank 3 of 7 in this publisher table
MMLU ProKnowledge87.97%Vals AInot published
MMMU ProMultimodal reasoning88.15%Vals AI
80.52%Artificial Analysis
not published
Vibe Code Bench v1.1Product coding84.96%Vals AInot published
Terminal-Bench Science 0.1Agentic science
2.86%Vals AI
not published
ProgramBenchAgentic coding2.00%Vals AI
17.5%ProgramBenchDeepSeek · 17 October 2026Rank 6 of 6 in this publisher table
17.5%ProgramBench (Almost Solved)Z.ai · 14 August 2026Rank 4 of 7 in this publisher table
17.5%ProgramBench (Almost@1)DeepSeek · 10 September 2026Rank 5 of 6 in this publisher table
ProofBench v1.1Mathematics87.00%Vals AInot published
LegalBenchProfessional knowledge86.02%Vals AI
86.2%LegalBenchZ.ai · 12 October 2026Rank 3 of 5 in this publisher table
Harvey's Legal Agent BenchmarkProfessional knowledge10.83%Vals AI
12.9%Harvey's Legal Agent BenchmarkZ.ai · 12 October 2026Rank 2 of 5 in this publisher table
Vals Multimodal IndexComposite73.42%Vals AInot published
Vals IndexComposite50.70%Vals AInot published
BioMysteryBenchAgentic science71.48%Vals AInot published
Code MigrationAgentic coding16.10%Vals AInot published
CyberBench v1.1Security75.24%Vals AInot published
EMBFinancial modelling66.40%Vals AInot published
Finance Agent (v2)Financial analysis54.36%Vals AInot published
IOICompetitive programming48.94%Vals AInot published
Legal Research BenchProfessional knowledge44.23%Vals AInot published
MedCodeHealthcare administration48.88%Vals AInot published
MedScribeHealthcare administration87.96%Vals AInot published
Public Benefits Bench v1.1Public services68.27%Vals AInot published
SAGEEducation54.26%Vals AInot published
Tax Agent BenchProfessional knowledge68.67%Vals AInot published
AutomationBenchBusiness workflowsno published result
46.7%AutomationBench (v1.0.6)DeepSeek · 17 October 2026Rank 5 of 6 in this publisher table
46.7%AutomationBench (v1.0.6)Z.ai · 14 August 2026Rank 2 of 8 in this publisher table
46.7%AutomationBench (Pass@1)DeepSeek · 10 September 2026Rank 4 of 7 in this publisher table
DeepSWE v1.1Software engineering68.51%Datacurve
67.5%DeepSWE v1.1DeepSeek · 17 October 2026Rank 5 of 6 in this publisher table
68.0%DeepSWE v1.1Reflection AI · 5 October 2026Rank 2 of 6 in this publisher table
67.5%DeepSWE (v1.1)Z.ai · 14 August 2026Rank 3 of 8 in this publisher table
67.5%DeepSWE v1.1 (Resolved)DeepSeek · 10 September 2026Rank 4 of 7 in this publisher table
GDPval-AA v2.1Knowledge work1,539.99Artificial Analysisnot published
AA-Briefcase v1.1Knowledge work1,500.5Artificial Analysisnot published
ChartographyVisual chart recognitionno published result
26.6%Chartography (no tools)Anthropic · 7 October 2026Rank 7 of 10 in this publisher table
68.1%Chartography (w/tools)DeepSeek · 17 October 2026Rank 5 of 5 in this publisher table
68.1%Chartography w/ tools (Pass@1)DeepSeek · 10 September 2026Rank 4 of 4 in this publisher table
Humanity's Last ExamMultidisciplinary reasoning46.90%Artificial Analysis
43.5%HLE (w/o tools)DeepSeek · 17 October 2026Rank 3 of 5 in this publisher table
46.9%HLE no toolsReflection AI · 5 October 2026Rank 1 of 8 in this publisher table
59.8%HLE w/ ToolsZ.ai · 14 August 2026Rank 5 of 8 in this publisher table
43.5%HLE (Pass@1)DeepSeek · 10 September 2026Rank 3 of 7 in this publisher table
59.8%HLE w/ tools (Pass@1)DeepSeek · 10 September 2026Rank 5 of 6 in this publisher table
Agents' Last ExamComputer useno published result
27.6%Agents' Last ExamDeepSeek · 17 October 2026Rank 5 of 6 in this publisher table
27.6%Agents' Last Exam (ALE-CLI)Z.ai · 14 August 2026Rank 3 of 8 in this publisher table
27.6%Agents' Last Exam (Pass@1)DeepSeek · 10 September 2026Rank 4 of 7 in this publisher table
Artificial Analysis Intelligence IndexComposite44Artificial Analysisnot published
Terminal-Bench 3.0Agentic codingno published result
17.4%Terminal Bench 3.0Z.ai · 14 August 2026Rank 5 of 6 in this publisher table
17.7%Terminal-Bench 3.0 (Pass@1)DeepSeek · 10 September 2026Rank 5 of 7 in this publisher table
NL2Repo-BenchAgentic codingno published result
58.0%NL2Repo-BenchDeepSeek · 17 October 2026Rank 4 of 6 in this publisher table
58.0%NL2RepoZ.ai · 14 August 2026Rank 3 of 6 in this publisher table
58.0%NL2Repo-Bench (Score)DeepSeek · 10 September 2026Rank 4 of 7 in this publisher table
CyberGymSecurityno published result
80.0%CyberGymZ.ai · 14 August 2026Rank 5 of 8 in this publisher table
80.0%CyberGym (Pass@1)DeepSeek · 10 September 2026Rank 5 of 6 in this publisher table
ExploitGymSecurityno published result
36ExploitGym (2h)Z.ai · 14 August 2026Rank 5 of 7 in this publisher table
70ExploitGym (6h)Z.ai · 14 August 2026Rank 5 of 7 in this publisher table
MathArena ApexMathematicsno published result
65.6%MathArena Apex (Pass@1)DeepSeek · 10 September 2026Rank 1 of 4 in this publisher table
BabyVisionMultimodalno published result
85.7%BabyVision (w/tools)DeepSeek · 17 October 2026Rank 5 of 5 in this publisher table
85.7%BabyVision w/tools (Pass@1)DeepSeek · 10 September 2026Rank 4 of 4 in this publisher table
ZeroBench (main set)Multimodalno published result
41.0%ZeroBench-main (w/tools)DeepSeek · 17 October 2026Rank 5 of 5 in this publisher table
41.0%ZeroBench-main w/tools (Pass@5)DeepSeek · 10 September 2026Rank 4 of 4 in this publisher table
SWE-MarathonAgentic codingno published result
48.1%SWE-Marathon (v1.1)Z.ai · 14 August 2026Rank 2 of 6 in this publisher table
PostTrainBenchAI researchno published result
32.0%PostTrainBenchZ.ai · 14 August 2026Rank 5 of 6 in this publisher table
ExploitBenchSecurityno published result
32.2%ExploitBenchZ.ai · 14 August 2026Rank 5 of 7 in this publisher table
Toolathlon VerifiedAgentic toolingno published result
76.5%Toolathlon VerifiedZ.ai · 14 August 2026Rank 1 of 8 in this publisher table
GDPval-AA v2Knowledge workno published result
1,682GDPval-AA v2Z.ai · 14 August 2026Rank 5 of 8 in this publisher table
Independent column: the suite named under each figure, one harness per suite. Publisher column: each lab’s own table, reproduced on the benchmark’s page. Publisher ranks compare reported scores in the same table row only; ties share a rank and unreported cells are excluded.

Sources

Read on 17 October 2026.