Moonshot AI model product
Kimi K3
Kimi K3 is Moonshot AI’s open Mixture-of-Experts language model for coding, knowledge work, and reasoning.
Updated Oct 23, 2026. Default version: Kimi K3
LLMBoard Score86.3Kimi K3
Coverage80%26 benchmark families
Context window1MTokens
Official input price$3Moonshot AI API
Kimi K3 Capability Profile
This profile uses the model's current scored version. Arena ratings and prices are shown separately.
Kimi K3 Benchmark Results
Benchmark scores for Kimi K3.
Sort by
Benchmark | Score | Rank | Participants | Percentile | Evidence | Evaluated |
|---|
| BenchmarkDECK-Bench | Score73.50% | Rank01 | Participants1 | Percentile100.00% | EvidenceC | Evaluated |
| BenchmarkKimi Code Bench v2 | Score72.90% | Rank01 | Participants2 | Percentile100.00% | EvidenceC | Evaluated |
| BenchmarkMathVision | Score97.80% | Rank01 | Participants38 | Percentile100.00% | EvidenceC | Evaluated |
| BenchmarkMLS-Bench Lite | Score48.30% | Rank01 | Participants3 | Percentile100.00% | EvidenceC | Evaluated |
| BenchmarkOmniDocBench | Score91.10% | Rank01 | Participants2 | Percentile100.00% | EvidenceC | Evaluated |
| BenchmarkSpreadsheetBench 2 | Score34.80% | Rank01 | Participants2 | Percentile100.00% | EvidenceC | Evaluated |
| BenchmarkLM Arena Agent Tool Hallucination | Score0.71% | Rank01 | Participants44 | Percentile100.00% | EvidenceA | Evaluated |
| BenchmarkAA LCR v1.1 | Score88.67% | Rank01 | Participants193 | Percentile100.00% | EvidenceB | EvaluatedN/A |
| BenchmarkAA-Briefcase | Score1,548.00 points | Rank02 | Participants4 | Percentile66.67% | EvidenceB | Evaluated |
| BenchmarkCharXiv-R | Score91.30% | Rank02 | Participants60 | Percentile98.31% | EvidenceC | Evaluated |
| BenchmarkDeepSearchQA | Score95.00% | Rank02 | Participants13 | Percentile91.67% | EvidenceC | Evaluated |
| BenchmarkFrontierSWE | Score81.20% | Rank02 | Participants17 | Percentile93.75% | EvidenceC | Evaluated |
| BenchmarkMMMU-Pro (with tools) | Score83.40% | Rank02 | Participants4 | Percentile66.67% | EvidenceC | Evaluated |
| BenchmarkPerceptionBench | Score58.50% | Rank02 | Participants2 | Percentile0.00% | EvidenceC | Evaluated |
| BenchmarkAPEX-Agents | Score37.60% | Rank03 | Participants11 | Percentile80.00% | EvidenceC | Evaluated |
| BenchmarkBrowseComp | Score91.20% | Rank03 | Participants70 | Percentile97.10% | EvidenceC | Evaluated |
| BenchmarkDeepSWE | Score67.50% | Rank03 | Participants13 | Percentile83.33% | EvidenceC | Evaluated |
| BenchmarkMCP Atlas | Score84.20% | Rank03 | Participants39 | Percentile94.74% | EvidenceC | Evaluated |
| BenchmarkWorldVQA | Score51.00% | Rank03 | Participants7 | Percentile66.67% | EvidenceC | Evaluated |
| BenchmarkBabyVision | Score85.70% | Rank04 | Participants16 | Percentile80.00% | EvidenceC | Evaluated |
| BenchmarkProgram Bench | Score77.80% | Rank04 | Participants18 | Percentile82.35% | EvidenceC | Evaluated |
| BenchmarkLiveBench reasoning (2026-06-25) | Score90.67 score | Rank04 | Participants43 | Percentile92.86% | EvidenceB | EvaluatedN/A |
| BenchmarkPostTrainBench | Score36.60% | Rank05 | Participants9 | Percentile50.00% | EvidenceC | Evaluated |
| BenchmarkZEROBench | Score0.41 points | Rank05 | Participants15 | Percentile71.43% | EvidenceC | Evaluated |
| BenchmarkAA SciCode Subtasks | Score59.49% | Rank05 | Participants89 | Percentile95.45% | EvidenceB | EvaluatedN/A |
| BenchmarkTerminal-Bench 2.1 | Score88.30% | Rank06 | Participants48 | Percentile89.36% | EvidenceC | Evaluated |
| BenchmarkOfficeQA Pro | Score63.30% | Rank07 | Participants12 | Percentile45.45% | EvidenceC | Evaluated |
| BenchmarkSWE-Marathon | Score42.00% | Rank07 | Participants10 | Percentile33.33% | EvidenceC | Evaluated |
| BenchmarkToolathlon | Score73.20% | Rank07 | Participants42 | Percentile85.37% | EvidenceC | Evaluated |
| BenchmarkGPQA | Score93.50% | Rank08 | Participants256 | Percentile97.25% | EvidenceC | Evaluated |
Kimi K3 Arena Results
Preference and agent-evaluation results for the default version.
Sort by
Arena | Category | Rank | Rating / score | Votes | Observations | Result date |
|---|
| Arenaagent tool hallucination | Categoryoverall | Rank01 | Rating / score0.01 | VotesN/A | Observations12.1M | Result date |
| Arenatext factuality | Categoryjapanese | Rank03 | Rating / score1,497.25 | Votes319 | ObservationsN/A | Result date |
| Arenatext factuality | Categoryindustry medicine and healthcare | Rank05 | Rating / score1,517.64 | Votes1,933 | ObservationsN/A | Result date |
| Arenatext style control | Categoryindustry legal and government | Rank05 | Rating / score1,510.32 | Votes2,682 | ObservationsN/A | Result date |
| Arenatext style control | Categoryindustry life and physical and social science | Rank05 | Rating / score1,524.17 | Votes5,364 | ObservationsN/A | Result date |
| Arenatext style control | Categoryindustry medicine and healthcare | Rank06 | Rating / score1,513.75 | Votes2,339 | ObservationsN/A | Result date |
| Arenatext style control | Categoryjapanese | Rank06 | Rating / score1,506.92 | Votes662 | ObservationsN/A | Result date |
| Arenatext factuality | Categoryindustry legal and government | Rank07 | Rating / score1,509.08 | Votes2,246 | ObservationsN/A | Result date |
| Arenatext style control | Categorygerman | Rank07 | Rating / score1,505.65 | Votes671 | ObservationsN/A | Result date |
| Arenatext style control | Categoryhard prompts english | Rank07 | Rating / score1,517.01 | Votes8,333 | ObservationsN/A | Result date |
| Arenatext style control | Categorymulti turn | Rank07 | Rating / score1,501.14 | Votes5,244 | ObservationsN/A | Result date |
| Arenaagent task outcome explicit | Categoryoverall | Rank08 | Rating / score0.06 | VotesN/A | Observations92.2K | Result date |
| Arenatext | Categoryindustry legal and government | Rank08 | Rating / score1,502.77 | Votes2,682 | ObservationsN/A | Result date |
| Arenatext style control | Categorycoding | Rank08 | Rating / score1,539.45 | Votes8,643 | ObservationsN/A | Result date |
| Arenatext style control | Categorykorean | Rank08 | Rating / score1,472.18 | Votes717 | ObservationsN/A | Result date |
| Arenatext style control | Categorylonger query | Rank08 | Rating / score1,503.11 | Votes14,884 | ObservationsN/A | Result date |
| Arenawebdev | Categoryoverall | Rank08 | Rating / score1,655.29 | Votes19,592 | ObservationsN/A | Result date |
| Arenawebdev | Categorywebdev | Rank08 | Rating / score1,655.29 | Votes19,592 | ObservationsN/A | Result date |
| Arenawebdev | Categorywebdev-html | Rank08 | Rating / score1,644.07 | Votes2,678 | ObservationsN/A | Result date |
| Arenawebdev | Categorywebdev-react | Rank08 | Rating / score1,670.46 | Votes12,988 | ObservationsN/A | Result date |
| Arenatext | Categoryindustry life and physical and social science | Rank09 | Rating / score1,508.41 | Votes5,364 | ObservationsN/A | Result date |
| Arenatext | Categoryjapanese | Rank09 | Rating / score1,493.59 | Votes662 | ObservationsN/A | Result date |
| Arenatext factuality | Categoryindustry mathematical | Rank09 | Rating / score1,488.44 | Votes1,498 | ObservationsN/A | Result date |
| Arenatext style control | Categorychinese | Rank09 | Rating / score1,541.37 | Votes2,501 | ObservationsN/A | Result date |
| Arenatext style control | Categoryexpert | Rank09 | Rating / score1,532.42 | Votes3,128 | ObservationsN/A | Result date |
| Arenatext style control | Categoryhard prompts | Rank09 | Rating / score1,516.42 | Votes21,953 | ObservationsN/A | Result date |
| Arenatext factuality | Categoryindustry life and physical and social science | Rank10 | Rating / score1,509.60 | Votes4,953 | ObservationsN/A | Result date |
| Arenatext style control | Categoryfrench | Rank10 | Rating / score1,510.30 | Votes1,305 | ObservationsN/A | Result date |
| Arenatext style control | Categoryindustry mathematical | Rank10 | Rating / score1,509.39 | Votes1,811 | ObservationsN/A | Result date |
| Arenatext | Categoryexpert | Rank11 | Rating / score1,520.39 | Votes3,128 | ObservationsN/A | Result date |
Kimi K3 Pricing
Official vendor API pricing appears first, followed by individual provider offers.
- Official API
- $3 input, $15 output per 1M
- Official provider
- Moonshot AI
- Lowest third-party
- From $0.239 input, $11.94 output per 1M via OpenRouter
- Tracked offerings
- 82
Sort by
Provider | Provider model ID | Region | Input / 1M | Output / 1M | Context | Updated |
|---|
| ProviderUmans AI Coding Plan | Provider model IDumans-kimi-k3 | Regionglobal | Input / 1MN/A | Output / 1MN/A | Context1M | Updated |
| ProviderNvidia | Provider model IDmoonshotai/kimi-k3 | Regionglobal | Input / 1MN/A | Output / 1MN/A | Context1M | Updated |
| ProviderSCNet Token Plan | Provider model IDKimi-K3 | Regionglobal | Input / 1MN/A | Output / 1MN/A | Context1M | Updated |
| ProviderSenseNova (China) | Provider model IDkimi-k3 | Regionglobal | Input / 1MN/A | Output / 1MN/A | Context1M | Updated |
| ProviderVolcengine Ark Coding Plan | Provider model IDkimi-k3 | Regionglobal | Input / 1MN/A | Output / 1MN/A | Context1M | Updated |
| ProviderKimi For Coding (kimi.ai) | Provider model IDk3 | Regionglobal | Input / 1MN/A | Output / 1MN/A | Context1M | Updated |
| ProviderKimi For Coding (kimi.com) | Provider model IDk3 | Regionglobal | Input / 1MN/A | Output / 1MN/A | Context1M | Updated |
| ProviderOpenRouter | Provider model IDmoonshotai/kimi-k3 | Regionglobal | Input / 1M$0.239 | Output / 1M$11.94 | Context1M | Updated |
| ProviderKenari | Provider model IDkimi-k3 | Regionglobal | Input / 1M$0.8571 | Output / 1M$8.57 | Context1M | Updated |
| ProviderNanoGPT | Provider model IDmoonshotai/kimi-k3 | Regionglobal | Input / 1M$1.2 | Output / 1M$5.5 | Context1M | Updated |
| Providerengy | Provider model IDkimi-k3 | Regionglobal | Input / 1M$1.95 | Output / 1M$9.75 | Context1.1M | Updated |
| Providerainetcafe | Provider model IDKimi-K3 | Regionglobal | Input / 1M$2.1 | Output / 1M$10.5 | Context262.1K | Updated |
| ProviderBoundless | Provider model IDkimi-k3 | Regionglobal | Input / 1M$2.3 | Output / 1M$11.4 | Context1M | Updated |
| ProviderIteraCompute | Provider model IDmoonshotai/kimi-k3 | Regionglobal | Input / 1M$2.35 | Output / 1M$11.76 | Context1M | Updated |
| ProviderVancine | Provider model IDkimi-k3 | Regionglobal | Input / 1M$2.4 | Output / 1M$12 | Context1M | Updated |
| ProviderWallaby | Provider model IDmoonshotai/kimi-k3 | Regionglobal | Input / 1M$2.7 | Output / 1M$13.5 | Context1M | Updated |
| ProviderSiliconFlow | Provider model IDmoonshotai/Kimi-K3 | Regionglobal | Input / 1M$2.7 | Output / 1M$13.5 | Context1M | Updated |
| ProviderAlibaba (China) | Provider model IDkimi-k3 | Regionglobal | Input / 1M$2.83 | Output / 1M$14.13 | Context1M | Updated |
| ProviderDeep Infra | Provider model IDmoonshotai/Kimi-K3 | Regionglobal | Input / 1M$2.85 | Output / 1M$14.25 | Context1M | Updated |
| ProviderMerge Gateway | Provider model IDmoonshot/kimi-k3 | Regionglobal | Input / 1M$2.9 | Output / 1M$14 | Context1M | Updated |
| ProviderMoonshot AI (China) | Provider model IDkimi-k3 | Regionglobal | Input / 1M$2.99 | Output / 1M$14.94 | Context1M | Updated |
| ProviderNeuralwatt | Provider model IDkimi-k3 | Regionglobal | Input / 1M$3 | Output / 1M$15 | Context1M | Updated |
| ProviderTokenGo | Provider model IDmoonshotai/kimi-k3 | Regionglobal | Input / 1M$3 | Output / 1M$15 | Context1M | Updated |
| ProviderOrcaRouter | Provider model IDkimi/kimi-k3 | Regionglobal | Input / 1M$3 | Output / 1M$15 | Context1M | Updated |
| ProviderCortecs | Provider model IDkimi-k3 | Regionglobal | Input / 1M$3 | Output / 1M$15 | Context1M | Updated |
| ProviderSynthetic | Provider model IDhf:moonshotai/Kimi-K3 | Regionglobal | Input / 1M$3 | Output / 1M$15 | Context524.3K | Updated |
| ProviderJalapeno Cloud | Provider model IDKimi-K3 | Regionglobal | Input / 1M$3 | Output / 1M$15 | Context1M | Updated |
| ProviderHugging Face | Provider model IDmoonshotai/Kimi-K3 | Regionglobal | Input / 1M$3 | Output / 1M$15 | Context1M | Updated |
| ProviderImpossibl | Provider model IDmoonshotai/kimi-k3 | Regionglobal | Input / 1M$3 | Output / 1M$15 | Context1M | Updated |
| ProviderMoonshot AI | Provider model IDkimi-k3 | Regionglobal | Input / 1M$3 | Output / 1M$15 | Context1M | Updated |
Kimi K3 Runtime Performance
Provider-specific output speed and catalog latency for Kimi K3. Runtime does not affect the capability score.
No runtime data
No provider-specific speed or latency record is linked to the default version yet.
Output Speed is generated output tokens received per second. Catalog latency is reported separately from observed provider TTFT.
Kimi K3 Specifications
Technical details for the model's default version.
- Version
- Kimi K3
- Released
- Jul 16, 2026
- Knowledge cutoff
- Unknown
- Parameters
- 2.8T
- Context window
- 1M
- Max output
- 1M
- Inputs
- image, text, video
- Outputs
- text
- Open weights
- Yes
- License
- Kimi K3 License
Kimi K3 Versions
Available versions of this model. The score column identifies the version used in the overall ranking.
Sort by
Version | Released | LLMBoard | Parameters | Context | Max output | Open weights | License |
|---|
| VersionKimi K3 | Released | LLMBoard86.25 | Parameters2.8T | Context1M | Max output1M | Open weightsYes | LicenseKimi K3 License |
Models similar to Kimi K3
Recommendations prioritize the same model type and family, then the closest LLMBoard score.
MA
Kimi K2.6
Moonshot AI
MA
Kimi K2.5
Moonshot AI
MA
Kimi K2.7 Code
Moonshot AI
MA
Kimi K2 Thinking
Moonshot AI
MA
Kimi K2
Moonshot AI
MA
Kimi k1.5
Moonshot AI
What is Kimi K3?
Key information about Kimi K3 and its available data.
8 trillion total parameters, 896 experts, and 16 active experts. It supports a 1M-token context window, native image and video understanding, tool calling, structured output, and always-on reasoning, and combines Kimi Delta Attention, Attention Residuals, and Stable LatentMoE.
Data as of 2026-10-22.
FAQ
Common questions about Kimi K3.
When was Kimi K3 released?
Kimi K3's default version was released on Jul 16, 2026.
How much does Kimi K3 cost?
Kimi K3's official API price is $3 per million input tokens and $15 per million output tokens via Moonshot AI. The lowest tracked third-party offer starts at $0.239 input and $11.94 output via OpenRouter.
Who created Kimi K3?
Kimi K3 was created by Moonshot AI.
What is the context window for Kimi K3?
The default version has a 1M token context window.
Is Kimi K3 open weight?
Yes. The default version is marked as open weight under Kimi K3 License.
How many API providers offer Kimi K3?
82 provider offerings are linked to the default version.