Skip to content

Kimi K3 VRAM requirements

How much VRAM Kimi K3 needs
INT4 weights: 1,454 GBWith 8K context: 1,455 GBWith 32K context: 1,456 GB
Parameters
2.8T
Active per token
104B
Longest context
1M
Released
Jun 2026
License
Kimi K3 License
Strongest at
Chat, Coding, Agents, Docs

How much VRAM Kimi K3 needs

Memory Kimi K3 needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
As released (INT4)1,454 GB1,455 GB1,456 GB1,459 GB
3-bit1,265 GB1,267 GB1,268 GB1,270 GB
2-bit971 GB973 GB973 GB976 GB
1.6-bit534 GB536 GB536 GB539 GB

One conversation, 16-bit context cache, runtime overhead included.

GPUs and Macs that run Kimi K3

The best way Kimi K3 runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Kimi K3 on RTX 4060 report
RTX 5060Too bigPDF: Kimi K3 on RTX 5060 report
RTX 3070Too bigPDF: Kimi K3 on RTX 3070 report
RTX 3080 10 GBToo bigPDF: Kimi K3 on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: Kimi K3 on RTX 3060 12 GB report
Arc B580Too bigPDF: Kimi K3 on Arc B580 report
RTX 4070Too bigPDF: Kimi K3 on RTX 4070 report
RTX 4070 SuperToo bigPDF: Kimi K3 on RTX 4070 Super report
RTX 5070Too bigPDF: Kimi K3 on RTX 5070 report
RTX 4060 Ti 16 GBToo bigPDF: Kimi K3 on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBToo bigPDF: Kimi K3 on RTX 5060 Ti 16 GB report
Arc A770 16 GBToo bigPDF: Kimi K3 on Arc A770 16 GB report
Radeon RX 9070 XTToo bigPDF: Kimi K3 on Radeon RX 9070 XT report
RTX 4070 Ti SuperToo bigPDF: Kimi K3 on RTX 4070 Ti Super report
RTX 4080 SuperToo bigPDF: Kimi K3 on RTX 4080 Super report
RTX 5070 TiToo bigPDF: Kimi K3 on RTX 5070 Ti report
RTX 5080Too bigPDF: Kimi K3 on RTX 5080 report
Radeon RX 7900 XTXToo bigPDF: Kimi K3 on Radeon RX 7900 XTX report
RTX 3090Too bigPDF: Kimi K3 on RTX 3090 report
RTX 4090Too bigPDF: Kimi K3 on RTX 4090 report
Radeon AI PRO R9700Too bigPDF: Kimi K3 on Radeon AI PRO R9700 report
RTX 5090Too bigPDF: Kimi K3 on RTX 5090 report
2× RTX 3090Too bigPDF: Kimi K3 on 2× RTX 3090 report
2× RTX 4090Too bigPDF: Kimi K3 on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GBToo bigPDF: Kimi K3 on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 AdaToo bigPDF: Kimi K3 on RTX 6000 Ada report
L40SToo bigPDF: Kimi K3 on L40S report
2× RTX 5090Too bigPDF: Kimi K3 on 2× RTX 5090 report
4× RTX 3090Too bigPDF: Kimi K3 on 4× RTX 3090 report
RTX PRO 6000 BlackwellToo bigPDF: Kimi K3 on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: Kimi K3 on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: Kimi K3 on Mac mini M4 24 GB report
Mac mini M4 32 GBToo bigPDF: Kimi K3 on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: Kimi K3 on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GBToo bigPDF: Kimi K3 on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GBToo bigPDF: Kimi K3 on Mac mini M4 Pro 64 GB report
M4 Max 128 GBToo bigPDF: Kimi K3 on M4 Max 128 GB report
M5 Max 128 GBToo bigPDF: Kimi K3 on M5 Max 128 GB report
M3 Ultra 96 GBToo bigPDF: Kimi K3 on M3 Ultra 96 GB report
M3 Ultra 256 GBToo bigPDF: Kimi K3 on M3 Ultra 256 GB report
M3 Ultra 512 GBToo bigPDF: Kimi K3 on M3 Ultra 512 GB report
M5 Ultra 512 GBToo bigPDF: Kimi K3 on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GBToo bigPDF: Kimi K3 on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GBToo bigPDF: Kimi K3 on Ryzen AI Max+ 395 128 GB report
RTX SparkToo bigPDF: Kimi K3 on RTX Spark report
DGX SparkToo bigPDF: Kimi K3 on DGX Spark report
Dual DGX SparkToo bigPDF: Kimi K3 on Dual DGX Spark report
DGX Station GB3001.6-bit, spills1M20 tok/s5.4 sPDF: Kimi K3 on DGX Station GB300 report
2× RTX PRO 6000 BlackwellToo bigPDF: Kimi K3 on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 BlackwellToo bigPDF: Kimi K3 on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell1.6-bit1M43 tok/s0.9 sPDF: Kimi K3 on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXMToo bigPDF: Kimi K3 on A100 80 GB SXM report
H100 SXMToo bigPDF: Kimi K3 on H100 SXM report
4× H200 NVLToo bigPDF: Kimi K3 on 4× H200 NVL report
8× H200 NVL2-bit1M58 tok/s0.9 sPDF: Kimi K3 on 8× H200 NVL report

Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.

How to run it

llama.cpp
llama-server -hf unsloth/Kimi-K3-GGUF
vLLM
vllm serve moonshotai/Kimi-K3

Questions

How much VRAM does Kimi K3 need?

Kimi K3 needs 1,454 GB for its weights as released (INT4), 1,455 GB with an 8K context and 1,456 GB with 32K.

What is the minimum VRAM for Kimi K3?

About 536 GB at 1.6-bit with an 8K context, the lowest precision we recommend for a model this size. The smallest machine here that runs it without spilling is the 8× RTX PRO 6000 Blackwell.

How big is Kimi K3?

2.8T parameters (104B active per token): 1,454 GB of weights as released (INT4).

Can Kimi K3 run on an RTX 4090?

Not on one card. The smallest machine here that runs it is the 8× RTX PRO 6000 Blackwell.

More models