Skip to content

Qwen3.5 397B A17B VRAM requirements

How much VRAM Qwen3.5 397B A17B needs
4-bit weights: 228 GBWith 8K context: 229 GBWith 32K context: 230 GB8-bit, 8K context: 401 GB
Parameters
403B
Active per token
17B
Longest context
256K
Released
Feb 2026
License
Apache 2.0
Strongest at
Agents, Docs

How much VRAM Qwen3.5 397B A17B needs

Memory Qwen3.5 397B A17B needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit751 GB753 GB754 GB756 GB
8-bit399 GB401 GB401 GB404 GB
6-bit308 GB310 GB310 GB313 GB
5-bit267 GB269 GB269 GB272 GB
4-bit228 GB229 GB230 GB233 GB
3-bit184 GB185 GB186 GB189 GB
2-bit141 GB142 GB143 GB146 GB
1.6-bit77.5 GB78.9 GB79.7 GB82.5 GB

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

GPUs and Macs that run Qwen3.5 397B A17B

The best way Qwen3.5 397B A17B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Qwen3.5 397B A17B on RTX 4060 report
RTX 5060Too bigPDF: Qwen3.5 397B A17B on RTX 5060 report
RTX 3070Too bigPDF: Qwen3.5 397B A17B on RTX 3070 report
RTX 3080 10 GBToo bigPDF: Qwen3.5 397B A17B on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: Qwen3.5 397B A17B on RTX 3060 12 GB report
Arc B580Too bigPDF: Qwen3.5 397B A17B on Arc B580 report
RTX 4070Too bigPDF: Qwen3.5 397B A17B on RTX 4070 report
RTX 4070 SuperToo bigPDF: Qwen3.5 397B A17B on RTX 4070 Super report
RTX 5070Too bigPDF: Qwen3.5 397B A17B on RTX 5070 report
RTX 4060 Ti 16 GBToo bigPDF: Qwen3.5 397B A17B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBToo bigPDF: Qwen3.5 397B A17B on RTX 5060 Ti 16 GB report
Arc A770 16 GBToo bigPDF: Qwen3.5 397B A17B on Arc A770 16 GB report
Radeon RX 9070 XTToo bigPDF: Qwen3.5 397B A17B on Radeon RX 9070 XT report
RTX 4070 Ti SuperToo bigPDF: Qwen3.5 397B A17B on RTX 4070 Ti Super report
RTX 4080 SuperToo bigPDF: Qwen3.5 397B A17B on RTX 4080 Super report
RTX 5070 TiToo bigPDF: Qwen3.5 397B A17B on RTX 5070 Ti report
RTX 5080Too bigPDF: Qwen3.5 397B A17B on RTX 5080 report
Radeon RX 7900 XTXToo bigPDF: Qwen3.5 397B A17B on Radeon RX 7900 XTX report
RTX 3090Too bigPDF: Qwen3.5 397B A17B on RTX 3090 report
RTX 4090Too bigPDF: Qwen3.5 397B A17B on RTX 4090 report
Radeon AI PRO R9700Too bigPDF: Qwen3.5 397B A17B on Radeon AI PRO R9700 report
RTX 5090Too bigPDF: Qwen3.5 397B A17B on RTX 5090 report
2× RTX 3090Too bigPDF: Qwen3.5 397B A17B on 2× RTX 3090 report
2× RTX 4090Too bigPDF: Qwen3.5 397B A17B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GBToo bigPDF: Qwen3.5 397B A17B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 AdaToo bigPDF: Qwen3.5 397B A17B on RTX 6000 Ada report
L40SToo bigPDF: Qwen3.5 397B A17B on L40S report
2× RTX 5090Too bigPDF: Qwen3.5 397B A17B on 2× RTX 5090 report
4× RTX 30901.6-bit256K108 tok/s1.0 sPDF: Qwen3.5 397B A17B on 4× RTX 3090 report
RTX PRO 6000 Blackwell1.6-bit256K153 tok/s0.8 sPDF: Qwen3.5 397B A17B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: Qwen3.5 397B A17B on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: Qwen3.5 397B A17B on Mac mini M4 24 GB report
Mac mini M4 32 GBToo bigPDF: Qwen3.5 397B A17B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: Qwen3.5 397B A17B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GBToo bigPDF: Qwen3.5 397B A17B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GBToo bigPDF: Qwen3.5 397B A17B on Mac mini M4 Pro 64 GB report
M4 Max 128 GB1.6-bit256K57 tok/s18 sPDF: Qwen3.5 397B A17B on M4 Max 128 GB report
M5 Max 128 GB1.6-bit256K61 tok/s5.1 sPDF: Qwen3.5 397B A17B on M5 Max 128 GB report
M3 Ultra 96 GBToo bigPDF: Qwen3.5 397B A17B on M3 Ultra 96 GB report
M3 Ultra 256 GB2-bit256K48 tok/s9.9 sPDF: Qwen3.5 397B A17B on M3 Ultra 256 GB report
M3 Ultra 512 GB6-bit256K29 tok/s9.9 sPDF: Qwen3.5 397B A17B on M3 Ultra 512 GB report
M5 Ultra 512 GB6-bit256K38 tok/s3.1 sPDF: Qwen3.5 397B A17B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GBToo bigPDF: Qwen3.5 397B A17B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB1.6-bit256K34 tok/s10 sPDF: Qwen3.5 397B A17B on Ryzen AI Max+ 395 128 GB report
RTX Spark1.6-bit256K47 tok/s3.6 sPDF: Qwen3.5 397B A17B on RTX Spark report
DGX Spark1.6-bit256K44 tok/s3.6 sPDF: Qwen3.5 397B A17B on DGX Spark report
Dual DGX Spark3-bit256K21 tok/s2.4 sPDF: Qwen3.5 397B A17B on Dual DGX Spark report
DGX Station GB3004-bit256K175 tok/s0.3 sPDF: Qwen3.5 397B A17B on DGX Station GB300 report
2× RTX PRO 6000 Blackwell2-bit256K113 tok/s0.5 sPDF: Qwen3.5 397B A17B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell6-bit256K67 tok/s0.3 sPDF: Qwen3.5 397B A17B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit256K55 tok/s0.2 sPDF: Qwen3.5 397B A17B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXMToo bigPDF: Qwen3.5 397B A17B on A100 80 GB SXM report
H100 SXMToo bigPDF: Qwen3.5 397B A17B on H100 SXM report
4× H200 NVL8-bit256K111 tok/s0.3 sPDF: Qwen3.5 397B A17B on 4× H200 NVL report
8× H200 NVL8-bit256K111 tok/s0.1 sPDF: Qwen3.5 397B A17B 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/Qwen3.5-397B-A17B-GGUF
vLLM
vllm serve Qwen/Qwen3.5-397B-A17B
MLX (Mac)
mlx_lm.chat --model mlx-community/Qwen3.5-397B-A17B-4bit

Questions

How much VRAM does Qwen3.5 397B A17B need?

Qwen3.5 397B A17B needs 228 GB for its weights at 4-bit, 229 GB with an 8K context and 230 GB with 32K. At 8-bit, 401 GB.

What is the minimum VRAM for Qwen3.5 397B A17B?

About 78.9 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 4× RTX 3090.

How big is Qwen3.5 397B A17B?

403B parameters (17B active per token): 751 GB of weights at 16-bit, 228 GB at 4-bit.

Can Qwen3.5 397B A17B run on an RTX 4090?

Not on one card. The smallest machine here that runs it is the 4× RTX 3090.

More models