Moonshot AI/Mixture of Experts

MKimi K3

Name: Kimi K3
Author: Moonshot AI

Kimi K3 — Moonshot's 2.8T MoE with 104B active, native multimodality and a 1M context. 69 of its 93 layers are Kimi Delta Attention and keep no growing KV cache; the other 24 use Gated MLA.

chatcodingreasoningmultilingualvisionmathagentictool_use

2779.93B

Parameters (104B active)

1024K

Context length

Benchmarks

Quantizations

2.1M

HF downloads

Architecture

MoE

Released

2026-07-16

Layers

KV Heads

Head Dim

Family

kimi

Quantization Options

Context length:

Quant	Bits	VRAM @ 16K	Quality
IQ2_XXS	2.38	827.8 GB 827.5 + 0.3 KV	low
IQ2_M	2.93	1018.9 GB 1018.6 + 0.3 KV	low
Q2_K	3.16	1098.9 GB 1098.6 + 0.3 KV	low
IQ3_XXS	3.25	1130.1 GB 1129.8 + 0.3 KV	low
IQ3_XS	3.5	1217.0 GB 1216.7 + 0.3 KV	low
Q3_K_S	3.64	1265.7 GB 1265.4 + 0.3 KV	low
IQ3_M	3.76	1307.4 GB 1307.1 + 0.3 KV	low
Q3_K_M	4	1390.8 GB 1390.5 + 0.3 KV	low
Q3_K_L	4.3	1495.0 GB 1494.7 + 0.3 KV	moderate
IQ4_XS	4.46	1550.6 GB 1550.3 + 0.3 KV	moderate
Q4_K_S	4.67	1623.6 GB 1623.3 + 0.3 KV	moderate
Q4_K_M	4.89	1700.0 GB 1699.7 + 0.3 KV	good
Q5_K_S	5.57	1936.3 GB 1936.0 + 0.3 KV	good
Q5_K_M	5.7	1981.5 GB 1981.2 + 0.3 KV	good
Q6_K	6.56	2280.3 GB 2280.0 + 0.3 KV	excellent
Q8_0	8.5	2954.5 GB 2954.2 + 0.3 KV	lossless
FP16	16	5560.7 GB 5560.3 + 0.3 KV	lossless

Select your GPU above to see speed estimates and compatibility for each quantization.

Too big for a single GPU — plan a multi-GPU deployment

Even the lightest quant needs ~828 GB. Size GPUs, replicas, TCO and scaling for a production setup. Open in Enterprise →

▸ READY TO RUN THIS?RENT BY THE HOUR

RENT A GPU AND RUN KIMI K3 NOW

Rent on Vast.ai →Or RunPod →

Spin up an A100 / H100 / 4090 in ~60s. Pay by the second. Cancel anytime.

Community Ratings

Loading ratings...

Benchmarks (6)

GPQA Diamond93.5

AA Long Context82.7

AA Coding76.2

AA Intelligence59.7

SciCode58.7

HLE46.9

Run this model

▸Easiest way to get started·Beginners

DOCS ↗

curl -fsSL https://ollama.com/install.sh | sh

$ollama run kimi:2780b-q4_K_M

Tag may need adjustment — check ollama.com/library/kimi for available tags.

▸ SETUP GUIDE

Auto-setup with fitmyllm CLI

Detects your GPU, recommends the best model, downloads it, and starts chatting — zero config. Benchmarks your speed and contributes anonymous data to improve predictions.

pip install fitmyllmthen run fitmyllmLearn more

Auto-detect GPULive tok/s in chatSpeed benchmarks9 inference engines

HuggingFace GGUF Downloads Build Hardware

Find the best GPU for Kimi K3

Build Hardware for Kimi K3

QUANT	BPW	VRAM	QUALITY
IQ2_XXS	2.38	827.5 GB	65%
IQ2_M	2.93	1018.6 GB	75%
Q2_K	3.16	1098.6 GB	78%
IQ3_XXS	3.25	1129.8 GB	82%
IQ3_XS	3.5	1216.7 GB	84%
Q3_K_S	3.64	1265.4 GB	85%
IQ3_M	3.76	1307.1 GB	86%
Q3_K_M	4	1390.5 GB	88%
Q3_K_L	4.3	1494.7 GB	90%
IQ4_XS	4.46	1550.3 GB	92%
Q4_K_S	4.67	1623.3 GB	93%
Q4_K_M	4.89	1699.7 GB	94%
Q5_K_S	5.57	1936.0 GB	96%
Q5_K_M	5.7	1981.2 GB	96%
Q6_K	6.56	2280.0 GB	97%
Q8_0	8.5	2954.2 GB	100%
FP16	16	5560.3 GB	100%