Moonshot AI model product
Kimi K3 is Moonshot AI's flagship open Mixture-of-Experts model for long-horizon coding, knowledge work, and reasoning.
Updated Aug 12, 2026. Default version: Kimi K3
Technical details for the model's default version.
This profile uses the model's current scored version. Arena ratings and prices are shown separately.
Benchmark scores for Kimi K3.
| AA-Briefcase | 1,548 points | 01 | 1 | 100.0% | B | |
| APEX-Agents | 37.6% | 01 | 7 | 100.0% | C | |
| AutomationBench | 30.8% | 01 | 8 | 100.0% | C | |
| BabyVision | 85.7% | 01 | 9 | 100.0% | C | |
| BrowseComp | 91.2% | 01 | 58 | 100.0% | C | |
| DECK-Bench | 73.5% | 01 | 1 | 100.0% | C | |
| DeepSearchQA | 95.0% | 01 | 9 | 100.0% | C | |
| Kimi Code Bench v2 | 72.9% | 01 | 2 | 100.0% | C | |
| MathVision | 97.8% | 01 | 32 | 100.0% | C | |
| MLS-Bench Lite | 48.3% | 01 | 3 | 100.0% | C | |
| OmniDocBench | 91.1% | 01 | 2 | 100.0% | C | |
| Program Bench | 77.8% | 01 | 5 | 100.0% | C | |
| SpreadsheetBench 2 | 34.8% | 01 | 1 | 100.0% | C | |
| SWE-Marathon | 42.0% | 01 | 3 | 100.0% | C | |
| ZEROBench | 0.41 points | 01 | 9 | 100.0% | C | |
| CharXiv-R | 91.3% | 02 | 48 | 97.9% | C | |
| FrontierSWE | 81.2% | 02 | 15 | 92.9% | C | |
| Job Bench | 52.9% | 02 | 2 | 0.0% | C | |
| MCP Atlas | 84.2% | 02 | 31 | 96.7% | C | |
| MMMU-Pro (with tools) | 83.4% | 02 | 4 | 66.7% | C | |
| PerceptionBench | 58.5% | 02 | 2 | 0.0% | C | |
| PostTrainBench | 36.6% | 02 | 5 | 75.0% | C | |
| Terminal-Bench 2.1 | 88.3% | 02 | 19 | 94.4% | C | |
| Toolathlon | 73.2% | 02 | 31 | 96.7% | C | |
| DeepSWE | 67.5% | 03 | 10 | 77.8% | C | |
| WorldVQA | 51.0% | 03 | 5 | 50.0% | C | |
| DeepSWE 1.1 | 69.0% | 04 | 20 | 84.2% | B | |
| OfficeQA Pro | 63.3% | 04 | 7 | 50.0% | C | |
| GPQA | 93.5% | 07 | 234 | 97.4% | C | |
| MMMU-Pro | 81.6% | 07 | 66 | 90.8% | C | |
| Humanity's Last Exam | 56.0% | 09 | 93 | 91.3% | C |
Preference and agent-evaluation results for the default version.
| webdev | overall | 02 | 1674.5 | 4,539 | N/A | |
| agent task outcome explicit | overall | 03 | 0.1 | N/A | 21.2K | |
| agent praise complaint | overall | 04 | 0.2 | N/A | 8.1K | |
| agent | overall | 05 | 0.1 | N/A | 2.2M | |
| agent steerability | overall | 10 | 0.1 | N/A | 25.6K | |
| agent tool hallucination | overall | 10 | 0.0 | N/A | 2.1M | |
| text style control | overall | 12 | 1487.3 | 11,438 | N/A | |
| text | overall | 16 | 1475.3 | 11,438 | N/A | |
| agent bash recovery steps | overall | 19 | 0.1 | N/A | 59.8K | |
| text factuality | overall | 21 | 1471.1 | 11,438 | N/A |
Provider-specific output speed and catalog latency for Kimi K3. Runtime does not affect the capability score.
| Novita | 39.843 tok/s | 26.673 s | 1M | 1M | |
| Fireworks | 18.262 tok/s | 8.115 s | 1M | 1M | |
| Moonshot AI | 10.366 tok/s | 9.33 s | 1M | 1M | |
| Together | 6.892 tok/s | 35.147 s | 1M | 1M |
Output Speed is generated output tokens received per second. Catalog latency is reported separately from observed provider TTFT.
Official vendor API pricing appears first, followed by individual provider offers.
| Kenari | kimi-k3 | global | N/A | N/A | 1M | |
| Umans AI Coding Plan | umans-kimi-k3 | global | N/A | N/A | 1M | |
| Kimi For Coding | k3 | global | N/A | N/A | 1M | |
| CrofAI | kimi-k3 | global | $2 | $8 | 1M | |
| Tinfoil | kimi-k3 | global | $2 | $6 | 256K | |
| Requesty | kimi-k3 | global | $2.25 | $11.25 | 1M | |
| NanoGPT | moonshotai/kimi-k3 | global | $2.5 | $13.5 | 1M | |
| DigitalOcean | kimi-k3 | global | $2.85 | $14.25 | 1M | |
| Deep Infra | moonshotai/Kimi-K3 | global | $2.85 | $14.25 | 1M | |
| Impossibl | moonshotai/kimi-k3 | global | $3 | $15 | 1M | |
| Fireworks AI | accounts/fireworks/models/kimi-k3 | global | $3 | $15 | 1M | |
| Weights & Biases | moonshotai/Kimi-K3 | global | $3 | $15 | 1M | |
| CrossModel | moonshot/kimi-k3 | global | $3 | $15 | 1M | |
| CoralBricks | kimi-k3 | global | $3 | $15 | 1M | |
| OpenCode Go | kimi-k3 | global | $3 | $15 | 1M | |
| Baseten | moonshotai/Kimi-K3 | global | $3 | $15 | 1M | |
| Nebius Token Factory | moonshotai/Kimi-K3 | global | $3 | $15 | 1M | |
| Vivgrid | kimi-k3 | global | $3 | $15 | 1M | |
| LLM Gateway | kimi-k3 | global | $3 | $15 | 1M | |
| OpenCode Zen | kimi-k3 | global | $3 | $15 | 1M | |
| GitHub Copilot | kimi-k3 | global | $3 | $15 | 1M | |
| Neon | kimi-k3 | global | $3 | $15 | 1M | |
| ClinePass | cline-pass/kimi-k3 | global | $3 | $15 | 1M | |
| ai& | moonshotai/kimi-k3 | global | $3 | $12.5 | 1M | |
| Pioneer | moonshotai/Kimi-K3 | global | $3 | $15 | 1M | |
| ZenMux | moonshotai/kimi-k3 | global | $3 | $15 | 1M | |
| EmpirioLabs AI | kimi-k3 | global | $3 | $15 | 1M | |
| Together AI | moonshotai/Kimi-K3 | global | $3 | $15 | 1M | |
| Moonshot AI (China) | kimi-k3 | global | $3 | $15 | 1M | |
| Modal | moonshotai/Kimi-K3 | global | $3 | $15 | 1M | |
| Berget.AI | moonshotai/Kimi-K3 | global | $3 | $15 | 327.7K | |
| Neuralwatt | kimi-k3 | global | $3 | $15 | 1M | |
| AIHubMix | kimi-k3 | global | $3 | $15 | 1M | |
| Umans AI | umans-kimi-k3 | global | $3 | $15 | 1M | |
| Ofox | moonshotai/kimi-k3 | global | $3 | $15 | 1M | |
| Synthetic | hf:moonshotai/Kimi-K3 | global | $3 | $15 | 524.3K | |
| TensorX | moonshotai/kimi-k3 | global | $3 | $15 | 1M | |
| Kilo Gateway | moonshotai/kimi-k3 | global | $3 | $15 | 1M | |
| Merge Gateway | moonshot/kimi-k3 | global | $3 | $15 | 1M | |
| Vercel AI Gateway | moonshotai/kimi-k3 | global | $3 | $15 | 1M | |
| NovitaAI | moonshotai/kimi-k3 | global | $3 | $15 | 1M | |
| OpenRouter | moonshotai/kimi-k3 | global | $3 | $15 | 1M | |
| Hugging Face | moonshotai/Kimi-K3 | global | $3 | $15 | 1M | |
| Cloudflare AI Gateway | moonshotai/kimi-k3 | global | $3 | $15 | 1M | |
| Perplexity Agent | moonshot-ai/kimi-k3 | global | $3 | $15 | 1M | |
| Moonshot AI | kimi-k3 | global | $3 | $15 | 1M | |
| Cortecs | kimi-k3 | global | $3 | $15 | 1M | |
| Charm Hyper | kimi-k3 | global | $3.27 | $16.33 | 1M | |
| Venice AI | kimi-k3 | global | $3.75 | $18.75 | 1M | |
| GreenPT | kimi-k3 | global | $3.76 | $18.81 | 1M | |
| LLM Gateway | kimi-k3-fast | global | $4.5 | $22.5 | 1M |
Official prices use only the vendor's configured official Provider and positive standard USD PAYG rates. Third-party offers remain explicitly labeled.
Available versions of this model. The score column identifies the version used in the overall ranking.
| Kimi K3 | 93.0 | 2.8T | 1M | 131.1K | Yes | Kimi K3 License |
Key information about Kimi K3 and its available data.
Kimi K3 is Moonshot AI's flagship open Mixture-of-Experts model for long-horizon coding, knowledge work, and reasoning. 8 trillion total parameters, activates 16 of 896 experts, and combines Kimi Delta Attention, Attention Residuals, and Stable LatentMoE.
The model supports a 1M-token context window, native image and video understanding, tool calling, structured output, and always-on reasoning with max thinking effort at launch.
Data as of 2026-08-11.
Open a comparison with the three ranked models immediately above and below this model.
Recommendations prioritize the same model type and family, then the closest LLMBoard score.
Common questions about Kimi K3.
Kimi K3's default version was released on Jul 16, 2026.
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 $2 input and $8 output via CrofAI.
Kimi K3 was created by Moonshot AI.
The default version has a 1M token context window.
Yes. The default version is marked as open weight under Kimi K3 License.
52 provider offerings are linked to the default version.
Nearby ranked alternatives include Claude Fable 5, GPT-5.6-Sol, Claude Mythos.