Request a pilot
‹ PublicAI Index

The LLM benchmark aggregator.

Kimi K3

Moonshot · 2.8T · open weights

Strongest in Finance (#6 of 151), weakest in Mathematics (#74 of 140). Above par in 23 of 28 scopes. Among the models it meets almost everywhere, it finishes behind Claude Fable 5.1 and Claude Fable 5 and ahead of Muse Spark 1.2 and Gemini 3 Pro.

§ 1 · Profile

What it is good at

Bars run from 50 — the average of the models each source lists — so right of the line is above par. The middle column is the gap to whoever leads that scope.

Professional58.9#9/168
Finance59.8#6/151
Biology research54.2#6/16
Legal60.2#11/151
Medical58.8#14/140
Cybersecurity42.8—/9✱
Core abilities60.1#9/204
Language61.7#10/57
General intelligence61.9#12/204
Data analysis58.9#15/57
Instruction following56.1#21/57
Human preference65.5#12/342
Human preference65.5#12/342
Coding58.5#14/165
Code generation58.2#17/77
Agentic coding58.6#19/157
Repository Q&A46—/0✱
Agents62.7#17/268
Knowledge work66.5#15/178
Workflow automation49.9—/0✱
Knowledge57#25/138
Academic knowledge58.4#22/123
Multimodal understanding35.9—/19✱
Reasoning56.7#27/178
Science61.8#13/122
Reasoning59.3#24/140
Mathematics51.5#74/140
Expert reasoning47.9—/0✱

§ 2 · Head to head

What it beats, and what beats it

The same models turn up scope after scope. Counted once: where both were placed, who finished higher. Bars run right for Kimi K3, left for the other.

GPT-4.1won 17 · lost 0
Kimi K2won 17 · lost 0
Mistral Medium 3.5won 17 · lost 0
Claude Opus 4won 17 · lost 0
Mimo V2.5 Prowon 17 · lost 0
GPT-4.1 Nanowon 18 · lost 0
GPT-4.1 Miniwon 18 · lost 0
Gemini 2.5 Flashwon 18 · lost 0
Mistral Large 3won 18 · lost 0
DeepSeek V3won 19 · lost 0
GPT-5.4 Miniwon 21 · lost 0
Grok 4.3won 22 · lost 0
Qwen3.6 Pluswon 22 · lost 0
Gemini 3.5 Flash Litewon 22 · lost 0
Kimi K2.6won 22 · lost 0
DeepSeek V4 Flashwon 26 · lost 1
Inklingwon 24 · lost 1
GPT-5.4 Nanowon 21 · lost 1
Claude Sonnet 4.6won 21 · lost 1
GPT-5.2won 20 · lost 1
GPT-6 Lunawon 19 · lost 1
GPT-4o Miniwon 18 · lost 1
Claude Sonnet 4won 18 · lost 1
Nemotron 3 Ultrawon 17 · lost 1
GPT OSS 120Bwon 17 · lost 1
GPT-5 Miniwon 17 · lost 1
Claude Haiku 4.5won 17 · lost 1
DeepSeek V3.2won 17 · lost 1
GPT-5won 17 · lost 1
Claude Sonnet 4.5won 17 · lost 1
O3 Miniwon 16 · lost 1
Grok 4won 16 · lost 1
Qwen3 235B A22B Instructwon 16 · lost 1
GLM-4.6won 16 · lost 1
Gemini 3.1 Flash Litewon 16 · lost 1
MiniMax M3won 23 · lost 2
Qwen3.8 27Bwon 21 · lost 2
GLM-5.3 Flashwon 20 · lost 2
Claude Opus 4.5won 19 · lost 2
Gemini 2.5 Prowon 17 · lost 2
DeepSeek R1won 16 · lost 2
Kimi K2.5won 16 · lost 2
GPT-5.6 Lunawon 22 · lost 3
GLM-5.2won 22 · lost 3
Qwen3.7 Maxwon 19 · lost 3
GPT-5.4won 19 · lost 3
Gemini 3.5 Flashwon 19 · lost 3
O4 Miniwon 15 · lost 3
Claude Sonnet 5won 20 · lost 4
Gemini 3.6 Flashwon 19 · lost 4
Grok 4.5won 18 · lost 4
DeepSeek V4 Prowon 21 · lost 5
O3won 14 · lost 4
Grok 4.7won 15 · lost 5
GPT-5.6 Terrawon 18 · lost 6
GLM-5.3won 19 · lost 7
Qwen3.8 Maxwon 15 · lost 7
Muse Spark 1.1won 15 · lost 7
DeepSeek V4.1 Flashwon 18 · lost 9
Gemini 3.1 Prowon 14 · lost 8
Gemini 3.7 Flashwon 14 · lost 8
Claude Opus 4.7won 14 · lost 8
Gemini 3.8 Flashwon 15 · lost 9
Claude Opus 4.6won 12 · lost 8
Claude Opus 4.8won 13 · lost 9
Gemini 3 Prowon 10 · lost 7
Muse Spark 1.2won 12 · lost 10
GPT-6 Solwon 10 · lost 10
GPT-5.5won 11 · lost 11
Grok 4.6won 11 · lost 13
GPT-5.6 Solwon 12 · lost 16
GPT-6 Astrawon 8 · lost 13
Muse Spark 1.3won 3 · lost 16
Claude Opus 5.5won 3 · lost 17
Claude Opus 5won 4 · lost 24
Claude Fable 5won 0 · lost 22
Claude Fable 5.1won 0 · lost 23

§ 3 · Sources

Where the numbers come from

9 publications, 46 figures. Every one links to the page it was read from.

LMArena Text ↗1 measureread 2026-09-26
LMArena Text 1488
Artificial Analysis ↗1 measureread 2026-09-26
Artificial Analysis Intelligence Index 44
GDPval-AA ↗1 measureread 2026-09-26
GDPval-AA 1524
AA-Briefcase ↗1 measureread 2026-09-26
AA-Briefcase 1505
ARC-AGI-2 ↗1 measureread 2026-09-26
ARC-AGI-2 60.4%
LiveBench ↗8 measuresread 2026-09-26
LiveBench 79.2LiveBench · Reasoning 90.7LiveBench · Coding 81.4LiveBench · Agentic Coding 62.2LiveBench · Mathematics 84.4LiveBench · Data Analysis 78.7LiveBench · Language 85.5LiveBench · Instruction Following 71.4
SimpleBench ↗1 measureread 2026-09-26
SimpleBench 60.7%
Vals.ai ↗16 measuresread 2026-09-26
Vals · Legal Research Bench 44.23%Vals · LegalBench 86.02%Vals · Harvey Legal Agent Benchmark 10.83%Vals · Finance Agent 54.36%Vals · CorpFin 71.56%Vals · TaxEval 75.72%Vals · MortgageTax 66.34%Vals · MedCode 48.88%Vals · MedScribe 87.96%Vals · BioMysteryBench 71.48%Vals · SWE-bench Verified 93.4% (Mini-SWE-agent)Vals · Vibe Code Bench 84.96% (OpenHands)Vals · Code Migration 16.1%Vals · GPQA Diamond 92.93%Vals · MMLU Pro 87.97%Vals · ProofBench 87%
GPQA Diamond (Pass@1) 92.9%HLE (Pass@1) 43.5%MathArena Apex (Pass@1) 65.6%Terminal-Bench 2.1 (Pass@1) 88.3%Terminal-Bench 3.0 (Pass@1) 17.7%Terminal-Bench 4.0 (Pass@1) 12.6%DeepSWE v1.1 (Resolved) 67.5%ProgramBench (Almost@1) 17.5%NL2Repo-Bench (Score) 58%CyberGym (Pass@1) 80%HLE w/ tools (Pass@1) 59.8%AutomationBench (Pass@1) 46.7%Agent's Last Exam (Pass@1) 27.6%Chartography w/ tools (Pass@1) 68.1%BabyVision w/ tools (Pass@1) 85.7%ZeroBench-main w/ tools (Pass@5) 41%

Badge

PublicAI Index badge for Kimi K3[![PublicAI Index](https://publicai.io/model-index/badge?model=kimi-k3)](https://publicai.io/model-index/m/kimi-k3)

✱ Placed by a one-off publication, not a board that re-ran the model. ✱✱ marks a figure the model’s own publisher printed, which counts for less again. Scores are 0–100 on the PublicAI Index scale, 50 = the average of the models each source lists. Scores belong to their publishers.

Snapshot 2026-09-26 · the full index · JSON API