Skip to content

Kimi K2 Instruct VRAM requirements

How much VRAM Kimi K2 Instruct needs
4-bit weights: 580 GBWith 8K context: 581 GBWith 32K context: 583 GB
Parameters
1T
Active per token
41.5B
Longest context
128K
Released
Jul 2025
License
Modified MIT
Strongest at
Not rated

Its licence (Modified MIT) is not a standard open-source one: check the terms before commercial use.

How much VRAM Kimi K2 Instruct needs

Memory Kimi K2 Instruct needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
As released (FP8)958 GB960 GB961 GB968 GB
6-bit784 GB785 GB787 GB793 GB
5-bit680 GB681 GB683 GB689 GB
4-bit580 GB581 GB583 GB589 GB
3-bit467 GB469 GB470 GB477 GB
2-bit358 GB360 GB362 GB368 GB
1.6-bit197 GB199 GB200 GB207 GB

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

GPUs and Macs that run Kimi K2 Instruct

The best way Kimi K2 Instruct runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Kimi K2 Instruct on RTX 4060 report
RTX 5060Too bigPDF: Kimi K2 Instruct on RTX 5060 report
RTX 3070Too bigPDF: Kimi K2 Instruct on RTX 3070 report
RTX 3080 10 GBToo bigPDF: Kimi K2 Instruct on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: Kimi K2 Instruct on RTX 3060 12 GB report
Arc B580Too bigPDF: Kimi K2 Instruct on Arc B580 report
RTX 4070Too bigPDF: Kimi K2 Instruct on RTX 4070 report
RTX 4070 SuperToo bigPDF: Kimi K2 Instruct on RTX 4070 Super report
RTX 5070Too bigPDF: Kimi K2 Instruct on RTX 5070 report
RTX 4060 Ti 16 GBToo bigPDF: Kimi K2 Instruct on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBToo bigPDF: Kimi K2 Instruct on RTX 5060 Ti 16 GB report
Arc A770 16 GBToo bigPDF: Kimi K2 Instruct on Arc A770 16 GB report
Radeon RX 9070 XTToo bigPDF: Kimi K2 Instruct on Radeon RX 9070 XT report
RTX 4070 Ti SuperToo bigPDF: Kimi K2 Instruct on RTX 4070 Ti Super report
RTX 4080 SuperToo bigPDF: Kimi K2 Instruct on RTX 4080 Super report
RTX 5070 TiToo bigPDF: Kimi K2 Instruct on RTX 5070 Ti report
RTX 5080Too bigPDF: Kimi K2 Instruct on RTX 5080 report
Radeon RX 7900 XTXToo bigPDF: Kimi K2 Instruct on Radeon RX 7900 XTX report
RTX 3090Too bigPDF: Kimi K2 Instruct on RTX 3090 report
RTX 4090Too bigPDF: Kimi K2 Instruct on RTX 4090 report
Radeon AI PRO R9700Too bigPDF: Kimi K2 Instruct on Radeon AI PRO R9700 report
RTX 5090Too bigPDF: Kimi K2 Instruct on RTX 5090 report
2× RTX 3090Too bigPDF: Kimi K2 Instruct on 2× RTX 3090 report
2× RTX 4090Too bigPDF: Kimi K2 Instruct on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GBToo bigPDF: Kimi K2 Instruct on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 AdaToo bigPDF: Kimi K2 Instruct on RTX 6000 Ada report
L40SToo bigPDF: Kimi K2 Instruct on L40S report
2× RTX 5090Too bigPDF: Kimi K2 Instruct on 2× RTX 5090 report
4× RTX 3090Too bigPDF: Kimi K2 Instruct on 4× RTX 3090 report
RTX PRO 6000 BlackwellToo bigPDF: Kimi K2 Instruct on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: Kimi K2 Instruct on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: Kimi K2 Instruct on Mac mini M4 24 GB report
Mac mini M4 32 GBToo bigPDF: Kimi K2 Instruct on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: Kimi K2 Instruct on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GBToo bigPDF: Kimi K2 Instruct on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GBToo bigPDF: Kimi K2 Instruct on Mac mini M4 Pro 64 GB report
M4 Max 128 GBToo bigPDF: Kimi K2 Instruct on M4 Max 128 GB report
M5 Max 128 GBToo bigPDF: Kimi K2 Instruct on M5 Max 128 GB report
M3 Ultra 96 GBToo bigPDF: Kimi K2 Instruct on M3 Ultra 96 GB report
M3 Ultra 256 GBToo bigPDF: Kimi K2 Instruct on M3 Ultra 256 GB report
M3 Ultra 512 GB2-bit80K26 tok/s26 sPDF: Kimi K2 Instruct on M3 Ultra 512 GB report
M5 Ultra 512 GB2-bit80K35 tok/s8.0 sPDF: Kimi K2 Instruct on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GBToo bigPDF: Kimi K2 Instruct on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GBToo bigPDF: Kimi K2 Instruct on Ryzen AI Max+ 395 128 GB report
RTX SparkToo bigPDF: Kimi K2 Instruct on RTX Spark report
DGX SparkToo bigPDF: Kimi K2 Instruct on DGX Spark report
Dual DGX Spark1.6-bit128K21 tok/s5.3 sPDF: Kimi K2 Instruct on Dual DGX Spark report
DGX Station GB3001.6-bit128K186 tok/s0.7 sPDF: Kimi K2 Instruct on DGX Station GB300 report
2× RTX PRO 6000 BlackwellToo bigPDF: Kimi K2 Instruct on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell2-bit44K62 tok/s0.7 sPDF: Kimi K2 Instruct on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell5-bit128K37 tok/s0.4 sPDF: Kimi K2 Instruct on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXMToo bigPDF: Kimi K2 Instruct on A100 80 GB SXM report
H100 SXMToo bigPDF: Kimi K2 Instruct on H100 SXM report
4× H200 NVL3-bit128K103 tok/s0.7 sPDF: Kimi K2 Instruct on 4× H200 NVL report
8× H200 NVLFP8128K62 tok/s0.3 sPDF: Kimi K2 Instruct 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-K2-Instruct-GGUF
vLLM
vllm serve moonshotai/Kimi-K2-Instruct
MLX (Mac)
mlx_lm.chat --model mlx-community/Kimi-K2-Instruct-4bit

Questions

How much VRAM does Kimi K2 Instruct need?

Kimi K2 Instruct needs 580 GB for its weights at 4-bit, 581 GB with an 8K context and 583 GB with 32K.

What is the minimum VRAM for Kimi K2 Instruct?

About 199 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 M3 Ultra 512 GB.

How big is Kimi K2 Instruct?

1T parameters (41.5B active per token): 958 GB of weights as released (FP8), 580 GB at 4-bit.

Can Kimi K2 Instruct run on an RTX 4090?

Not on one card. The smallest machine here that runs it is the M3 Ultra 512 GB.

More models