Kimi K3
Kimi K3 is an open-weights language model from Moonshot AI, released July 16, 2026. It has 2.8T parameters (dense). The context window is 1M tokens. On the Epoch Capabilities Index it scores 158.6, rank 24 of 256 rated models. API price: $3 input / $15 output per million tokens.
Today, we are introducing Kimi K3 — our most capable model. Kimi K3 is a 2.8T-parameter model built on our Kimi Delta Attention and Attention Residuals, with native vision capabilities and a 1-million-token context window.
Check it against your own PC →
Can you run Kimi K3 locally?
Even with 32 GB of VRAM and 32 GB of RAM it is too large to run at a usable speed on a single consumer GPU. Estimates assume 32 GB of DDR5 RAM, a 16K context and llama.cpp-style GGUF quantization. Enter your exact hardware in the live app.
| GPU memory | Verdict | Best quant | VRAM | RAM | tok/s |
|---|---|---|---|---|---|
| 6 GB VRAM | Too large | – | – | – | – |
| 8 GB VRAM | Too large | – | – | – | – |
| 12 GB VRAM | Too large | – | – | – | – |
| 16 GB VRAM | Too large | – | – | – | – |
| 24 GB VRAM | Too large | – | – | – | – |
| 32 GB VRAM | Too large | – | – | – | – |
Ready-made quantizations on Hugging Face: unsloth/Kimi-K3-GGUF · 6block/Kimi-K3-GGUF · Ryanchen911/Kimi-K3-Uncensored-GGUF · nvidia/Kimi-K3-NVFP4 · RedHatAI/Kimi-K3-NVFP4 · pipenetwork/Kimi-K3-REAP80-MLX-mxfp4-q8
Strengths by category
ModelTape's 0–100 fit on Epoch AI benchmark results.
| Category | Score | Based on |
|---|---|---|
| Reasoning | 19.4 | Chess Puzzles, SimpleBench, Mystery Game Puzzles |
| Math | 64.6 | ProofBench, OTIS Mock AIME 2024-2025, FrontierMath-Tiers-1-3-v2-Private, FrontierMath-Tier-4-v2-Private |
| Coding | 63.0 | MirrorCode, FrontierSWE, DeepSWE |
| Knowledge | 77.2 | GPQA diamond, SimpleQA Verified |
Benchmark results
| Benchmark (Epoch AI) | Score |
|---|---|
| OTIS Mock AIME | 97.2% |
| GPQA Diamond | 94.6% |
| ProofBench | 87.0% |
| FrontierMath (Tiers 1–3) | 76.0% |
| DeepSWE | 68.5% |
| MirrorCode | 54.4% |
| SimpleBench | 52.8% |
| SimpleQA Verified | 50.6% |
| FrontierMath-Tier-4-v2-Private | 39.0% |
| Chess Puzzles | 33.7% |
| FrontierSWE | 25.9% |
| Mystery Game Puzzles | 18.5% |
| Furniture Assembly | 6.0% |
Links
Announcement · Hugging Face: moonshotai/Kimi-K3 · Epoch AI benchmarks
Models with a similar score
Muse Wonder · Kimi K3.1 Pro · GPT-5.6 Terra · Claude Haiku 5.5 · GPT-5.4 Pro · GPT-5.5 · Claude Opus 4.8 · DeepSeek V4.1 Flash 1017
More from Moonshot AI
Kimi K3.1 Pro · Kimi K2.7 Code · Kimi K2.6 · Kimi K2.5 · Kimi K2 Thinking · Kimi Linear · Kimi K2 (Jul 2025) · Kimi-VL