Noometry

Moonshot AI, open weights

Kimi K3

Kimi K3 by Moonshot AI ranks 15th of 354 ranked models on the Noometry Index as of October 2026, with a score of 59.5. Its strongest category is writing & preference, where it ranks 4th. API pricing starts at $3 per million input tokens and $15 per million output tokens, with a 1.05M-token context window.

Last verified

Noometry rank
#15 of 354
Index score
59.5
Evidence
Confirmed 53 results
Provider
Moonshot AI
Released
July 16, 2026
Weights
Open weights
Reasoning
Yes
Context window
1.05M
Max output
1.05M
Input price
$3 / M
Output price
$15 / M
Blended price
$6 / M
Output speed
Not measured
Value
#185 of 219
Knowledge cutoff
Unknown
Input
text, image, video
Hugging Face
moonshotai/Kimi-K3

Category scores

Each category score combines every public result we have in that category.

  1. Coding 61.0
  2. Agentic & Tool Use 41.8
  3. Reasoning 63.0
  4. Math 74.2
  5. Knowledge 63.2
  6. Multimodal 37.8
  7. Multilingual 56.3
  8. Instruction Following 77.7
  9. Long Context 45.8
  10. Writing & Preference 76.6
CategoryScoreRankResults
Coding61.0#107
Agentic & Tool Use41.8#205
Reasoning63.0#1711
Math74.2#166
Knowledge63.2#213
Multimodal37.8#702
Multilingual56.3#211
Instruction Following77.7#141
Long Context45.8#291
Writing & Preference76.6#45

Strengths and weaknesses

Categories where Kimi K3 places highest and lowest among the models ranked in each, with its score against that category's median.

Strongest categories

CategoryScorevs medianRank
Writing & Preference76.6+22.8#4 of 312, top 2%
Coding61.0+22.3#10 of 340, top 3%
Instruction Following77.7+6.5#14 of 305, top 5%

Weakest categories

CategoryScorevs medianRank
Multimodal37.8−0.8#70 of 128, top 55%
Agentic & Tool Use41.8+11.5#20 of 154, top 13%
Long Context45.8+4.9#29 of 296, top 10%

Closest competitors

The models ranked just above and below Kimi K3. When scores are this close, price and speed are often the better way to choose.

ModelRankScoreBlended $/MSpeed
Gemini 3.8 Flash#1161.8$1.50—Compare
GPT-6 Sol#1261.8$4—Compare
Claude Opus 4.8#1360.7$1034Compare
Gemini 3.7 Flash#1459.8$1.50—Compare
GPT-5.4#1659.4$5.6312Compare
GPT-5.6 Terra#1759.2$4.5011Compare
GPT-5.4 Pro#1858.9$67.50—Compare
Claude Opus 4.7#1958.3$1033Compare

Sponsored placements are available on pages like this one. Advertise on Noometry

Benchmark results

Every published result we track, with its source. Bold rows are the ones used for ranking; where several exist we prefer independent runs over self-reported numbers.

Coding

BenchmarkScorePositionSettingSourceDate
DeepSWE68.5%#10 of 29, top 35%maxEpoch AI
FrontierCode44.2%#13 of 37, top 36%Epoch AI
LMArena WebDev1654#11 of 113, top 10%LMArena2026-10-08
FrontierSWE25.9%#11 of 18, top 62%maxEpoch AI
SciCode58.7%Epoch AI
SciCode51.2%lowEpoch AI
SciCode59.5%#9 of 121, top 8%maxEpoch AI
WeirdML82.6%#9 of 119, top 8%maxEpoch AI
LMArena Coding1508#12 of 294, top 5%LMArena2026-10-08
ALE-Bench1,524#16 of 105, top 16%maxEpoch AI

Agentic & Tool Use

BenchmarkScorePositionSettingSourceDate
APEX-Agents50.6%#25 of 49, top 52%Epoch AI
τ²-bench Banking37.1%#12 of 26, top 47%maxτ²-bench2026-08-04
PostTrainBench32%#5 of 11, top 46%Epoch AI
GBAEval48.3%#9 of 23, top 40%Epoch AI
GDP.pdf19%#21 of 36, top 59%maxEpoch AI
Vending-Bench 25,165#27 of 60, top 45%Epoch AI

Reasoning

BenchmarkScorePositionSettingSourceDate
ARC-AGI-255%highEpoch AI
ARC-AGI-212.4%lowEpoch AI
ARC-AGI-260.4%#29 of 83, top 35%maxEpoch AI
SimpleBench60.7%#23 of 77, top 30%maxEpoch AI
NYT Connections (extended)93.6%#10 of 91, top 11%Lech Mazur benchmarks
ARC-AGI-186.7%highEpoch AI
ARC-AGI-165.7%lowEpoch AI
ARC-AGI-194.5%#17 of 83, top 21%maxEpoch AI
CritPt3.1%lowEpoch AI
CritPt23.4%#19 of 134, top 15%maxEpoch AI
Chess Puzzles25%highEpoch AI2026-08-07
Chess Puzzles20%lowEpoch AI2026-08-07
Chess Puzzles39%#23 of 129, top 18%maxEpoch AI2026-07-16
LMArena Hard Prompts1496#12 of 297, top 5%LMArena2026-10-08
Mystery Game Puzzles12%lowEpoch AI2026-08-29
Mystery Game Puzzles26%#32 of 74, top 44%maxEpoch AI2026-07-29
DTBench91.2%#31 of 151, top 21%maxEpoch AI
LMCA52.7%#17 of 125, top 14%maxEpoch AI
Surface Evolver Bench95%#2 of 25, top 8%Epoch AI
Surface Evolver Bench93%maxEpoch AI
Epoch Capabilities Index157.45#15 of 213, top 8%Epoch AI2026-07-16
ForecastBench61.1#17 of 72, top 24%maxEpoch AI

Math

BenchmarkScorePositionSettingSourceDate
FrontierMath (Tiers 1-3)72.2%#22 of 81, top 28%maxEpoch AI2026-07-17
FrontierMath Tier 439%#23 of 63, top 37%maxEpoch AI2026-07-17
MathArena Final-Answer Competitions87.8%#3 of 29, top 11%thinkMathArena
OTIS Mock AIME 2024-202593.3%highEpoch AI2026-08-07
OTIS Mock AIME 2024-202568.9%lowEpoch AI2026-08-07
OTIS Mock AIME 2024-202597.2%#28 of 173, top 17%maxEpoch AI2026-07-16
ProofBench87%#9 of 77, top 12%Epoch AI
LMArena Math1491#17 of 285, top 6%LMArena2026-10-08

Knowledge

BenchmarkScorePositionSettingSourceDate
GPQA Diamond91.9%highEpoch AI2026-08-07
GPQA Diamond84.8%lowEpoch AI2026-08-07
GPQA Diamond93.1%#18 of 186, top 10%maxEpoch AI2026-07-16
SimpleQA Verified50.6%#24 of 77, top 32%maxEpoch AI2026-08-27
LMArena Expert1521#10 of 273, top 4%LMArena2026-10-08

Multimodal

BenchmarkScorePositionSettingSourceDate
Blueprint-Bench 229.5%#17 of 31, top 55%Epoch AI
Furniture Assembly34.2%#20 of 31, top 65%maxEpoch AI2026-09-11

Multilingual

BenchmarkScorePositionSettingSourceDate
LMArena Non-English1466#21 of 297, top 8%LMArena2026-10-08
LMArena Chinese1529#15 of 285, top 6%LMArena2026-10-08
LMArena French1491#19 of 223, top 9%LMArena2026-10-08
LMArena German1488#14 of 231, top 7%LMArena2026-10-08
LMArena Japanese1487#11 of 211, top 6%LMArena2026-10-08
LMArena Korean1458#14 of 213, top 7%LMArena2026-10-08
LMArena Russian1482#18 of 283, top 7%LMArena2026-10-08
LMArena Spanish1472#20 of 226, top 9%LMArena2026-10-08

Instruction Following

BenchmarkScorePositionSettingSourceDate
LMArena Instruction Following1483#11 of 298, top 4%LMArena2026-10-08

Long Context

BenchmarkScorePositionSettingSourceDate
LMArena Longer Query1494#11 of 291, top 4%LMArena2026-10-08

Writing & Preference

BenchmarkScorePositionSettingSourceDate
LMArena Text1476#20 of 297, top 7%LMArena2026-10-08
LMArena Creative Writing1454#25 of 295, top 9%LMArena2026-10-08
EQ-Bench Creative Writing2082#5 of 115, top 5%EQ-Bench
EQ-Bench 41339#3 of 28, top 11%EQ-Bench
LMArena Multi-Turn1488#13 of 295, top 5%LMArena2026-10-08

API pricing by provider

RouteInput $/MOutput $/MCached input $/MChecked
alibaba$3$15$0.302026-10-10
bedrock$3$15$0.302026-10-10
deepinfra$2.85$14.25$0.282026-10-10
fireworks$3$15$0.302026-10-10
moonshot$3$15$0.302026-10-10
openrouter$0.50$13.50$0.282026-10-10
together$3$15$0.302026-10-10

All Moonshot AI API prices → Estimate your cost →

Compare Kimi K3

Other Moonshot AI models

Frequently asked questions

How good is Kimi K3?

Kimi K3 by Moonshot AI ranks 15th of 354 ranked models on the Noometry Index as of October 2026, with a score of 59.5. Its strongest category is writing & preference, where it ranks 4th. API pricing starts at $3 per million input tokens and $15 per million output tokens, with a 1.05M-token context window.

How much does Kimi K3 cost?

Kimi K3 costs $3 per million input tokens and $15 per million output tokens on Moonshot AI's own API, with cached input at $0.30.

What is Kimi K3's context window?

Kimi K3 accepts up to 1.05M tokens of input and can write up to 1.05M tokens in one response.

Is Kimi K3 open source?

Yes. Kimi K3's weights are downloadable from Hugging Face (moonshotai/Kimi-K3); check the license for commercial terms.

What are Kimi K3's strengths and weaknesses?

Relative to other ranked models, Kimi K3 places best in writing & preference, coding, instruction following and lowest in multimodal, agentic & tool use, long context.

What is Kimi K3 best at?

Its best category is writing & preference, where it ranks 4th on Noometry.