Skip to content

Kimi Linear 48B A3B Instruct VRAM requirements

How much VRAM Kimi Linear 48B A3B Instruct needs
4-bit weights: 27.7 GBWith 4K context: 28.6 GB8-bit, 4K context: 49.5 GB
Parameters
49.1B
Active per token
3B
Longest context
4K
Released
Oct 2025
License
MIT
Strongest at
Not rated

How much VRAM Kimi Linear 48B A3B Instruct needs

Memory Kimi Linear 48B A3B Instruct needs at each precision and context length
PrecisionWeightsWith 4K
16-bit91.5 GB92.4 GB
8-bit48.6 GB49.5 GB
6-bit37.5 GB38.4 GB
5-bit32.5 GB33.4 GB
4-bit27.7 GB28.6 GB
3-bit22.4 GB23.3 GB
2-bit17.2 GB18.1 GB
1.6-bit9.4 GB10.3 GB

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

GPUs and Macs that run Kimi Linear 48B A3B Instruct

The best way Kimi Linear 48B A3B Instruct runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Kimi Linear 48B A3B Instruct on RTX 4060 report
RTX 5060Too bigPDF: Kimi Linear 48B A3B Instruct on RTX 5060 report
RTX 3070Too bigPDF: Kimi Linear 48B A3B Instruct on RTX 3070 report
RTX 3080 10 GBToo bigPDF: Kimi Linear 48B A3B Instruct on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: Kimi Linear 48B A3B Instruct on RTX 3060 12 GB report
Arc B580Too bigPDF: Kimi Linear 48B A3B Instruct on Arc B580 report
RTX 4070Too bigPDF: Kimi Linear 48B A3B Instruct on RTX 4070 report
RTX 4070 SuperToo bigPDF: Kimi Linear 48B A3B Instruct on RTX 4070 Super report
RTX 5070Too bigPDF: Kimi Linear 48B A3B Instruct on RTX 5070 report
RTX 4060 Ti 16 GBToo bigPDF: Kimi Linear 48B A3B Instruct on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBToo bigPDF: Kimi Linear 48B A3B Instruct on RTX 5060 Ti 16 GB report
Arc A770 16 GBToo bigPDF: Kimi Linear 48B A3B Instruct on Arc A770 16 GB report
Radeon RX 9070 XTToo bigPDF: Kimi Linear 48B A3B Instruct on Radeon RX 9070 XT report
RTX 4070 Ti SuperToo bigPDF: Kimi Linear 48B A3B Instruct on RTX 4070 Ti Super report
RTX 4080 SuperToo bigPDF: Kimi Linear 48B A3B Instruct on RTX 4080 Super report
RTX 5070 TiToo bigPDF: Kimi Linear 48B A3B Instruct on RTX 5070 Ti report
RTX 5080Too bigPDF: Kimi Linear 48B A3B Instruct on RTX 5080 report
Radeon RX 7900 XTX3-bit4K198 tok/s0.8 sPDF: Kimi Linear 48B A3B Instruct on Radeon RX 7900 XTX report
RTX 30903-bit4K255 tok/s0.5 sPDF: Kimi Linear 48B A3B Instruct on RTX 3090 report
RTX 40903-bit4K266 tok/s0.2 sPDF: Kimi Linear 48B A3B Instruct on RTX 4090 report
Radeon AI PRO R97004-bit4K137 tok/s0.5 sPDF: Kimi Linear 48B A3B Instruct on Radeon AI PRO R9700 report
RTX 50904-bit4K322 tok/s0.2 sPDF: Kimi Linear 48B A3B Instruct on RTX 5090 report
2× RTX 30906-bit4K184 tok/s0.3 sPDF: Kimi Linear 48B A3B Instruct on 2× RTX 3090 report
2× RTX 40906-bit4K194 tok/s0.1 sPDF: Kimi Linear 48B A3B Instruct on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB6-bit4K234 tok/s0.3 sPDF: Kimi Linear 48B A3B Instruct on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada6-bit4K187 tok/s0.2 sPDF: Kimi Linear 48B A3B Instruct on RTX 6000 Ada report
L40S6-bit4K174 tok/s0.3 sPDF: Kimi Linear 48B A3B Instruct on L40S report
2× RTX 50908-bit4K238 tok/s0.1 sPDF: Kimi Linear 48B A3B Instruct on 2× RTX 5090 report
4× RTX 30908-bit4K153 tok/s0.1 sPDF: Kimi Linear 48B A3B Instruct on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit4K238 tok/s0.2 sPDF: Kimi Linear 48B A3B Instruct on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: Kimi Linear 48B A3B Instruct on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: Kimi Linear 48B A3B Instruct on Mac mini M4 24 GB report
Mac mini M4 32 GBToo bigPDF: Kimi Linear 48B A3B Instruct on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: Kimi Linear 48B A3B Instruct on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB5-bit4K74 tok/s5.4 sPDF: Kimi Linear 48B A3B Instruct on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB6-bit4K67 tok/s5.4 sPDF: Kimi Linear 48B A3B Instruct on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit4K86 tok/s2.7 sPDF: Kimi Linear 48B A3B Instruct on M4 Max 128 GB report
M5 Max 128 GB8-bit4K93 tok/s0.8 sPDF: Kimi Linear 48B A3B Instruct on M5 Max 128 GB report
M3 Ultra 96 GB8-bit4K102 tok/s1.5 sPDF: Kimi Linear 48B A3B Instruct on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit4K102 tok/s1.5 sPDF: Kimi Linear 48B A3B Instruct on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit4K102 tok/s1.5 sPDF: Kimi Linear 48B A3B Instruct on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit4K126 tok/s0.5 sPDF: Kimi Linear 48B A3B Instruct on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB6-bit4K53 tok/s1.6 sPDF: Kimi Linear 48B A3B Instruct on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit4K43 tok/s1.6 sPDF: Kimi Linear 48B A3B Instruct on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit4K59 tok/s1.0 sPDF: Kimi Linear 48B A3B Instruct on RTX Spark report
DGX Spark8-bit4K54 tok/s1.1 sPDF: Kimi Linear 48B A3B Instruct on DGX Spark report
Dual DGX Spark8-bit4K54 tok/s0.6 sPDF: Kimi Linear 48B A3B Instruct on Dual DGX Spark report
DGX Station GB3008-bit4K441 tok/s0.1 sPDF: Kimi Linear 48B A3B Instruct on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit4K238 tok/s0.1 sPDF: Kimi Linear 48B A3B Instruct on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit4K238 tok/s0.1 sPDF: Kimi Linear 48B A3B Instruct on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit4K238 tok/s0.1 sPDF: Kimi Linear 48B A3B Instruct on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit4K257 tok/s0.3 sPDF: Kimi Linear 48B A3B Instruct on A100 80 GB SXM report
H100 SXM8-bit4K334 tok/s0.1 sPDF: Kimi Linear 48B A3B Instruct on H100 SXM report
4× H200 NVL8-bit4K387 tok/s0.1 sPDF: Kimi Linear 48B A3B Instruct on 4× H200 NVL report
8× H200 NVL8-bit4K387 tok/s0.1 sPDF: Kimi Linear 48B A3B 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 bartowski/moonshotai_Kimi-Linear-48B-A3B-Instruct-GGUF
vLLM
vllm serve moonshotai/Kimi-Linear-48B-A3B-Instruct
MLX (Mac)
mlx_lm.chat --model mlx-community/Kimi-Linear-48B-A3B-Instruct-4bit

Questions

How much VRAM does Kimi Linear 48B A3B Instruct need?

Kimi Linear 48B A3B Instruct needs 27.7 GB for its weights at 4-bit, 28.6 GB with an 4K context and 28.6 GB with 4K. At 8-bit, 49.5 GB.

What is the minimum VRAM for Kimi Linear 48B A3B Instruct?

About 23.3 GB at 3-bit with an 4K context, the lowest precision we recommend for a model this size. The smallest machine here that runs it without spilling is the Radeon RX 7900 XTX.

How big is Kimi Linear 48B A3B Instruct?

49.1B parameters (3B active per token): 91.5 GB of weights at 16-bit, 27.7 GB at 4-bit.

Can Kimi Linear 48B A3B Instruct run on an RTX 4090?

Yes. At 3-bit it holds 4K of context and writes about 266 tokens a second.

More models