Skip to content

Qwen3 235B A22B Instruct 2507 VRAM requirements

How much VRAM Qwen3 235B A22B Instruct 2507 needs
4-bit weights: 133 GBWith 8K context: 135 GBWith 32K context: 139 GB8-bit, 8K context: 235 GB
Parameters
235B
Active per token
22B
Longest context
256K
Released
Jul 2025
License
Apache 2.0
Strongest at
Not rated

How much VRAM Qwen3 235B A22B Instruct 2507 needs

Memory Qwen3 235B A22B Instruct 2507 needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit438 GB440 GB445 GB462 GB
8-bit233 GB235 GB239 GB257 GB
6-bit180 GB182 GB186 GB204 GB
5-bit156 GB158 GB162 GB180 GB
4-bit133 GB135 GB139 GB157 GB
3-bit107 GB109 GB114 GB131 GB
2-bit82.1 GB84.4 GB88.8 GB106 GB
1.6-bit45.2 GB47.5 GB51.9 GB69.5 GB

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

GPUs and Macs that run Qwen3 235B A22B Instruct 2507

The best way Qwen3 235B A22B Instruct 2507 runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Qwen3 235B A22B Instruct 2507 on RTX 4060 report
RTX 5060Too bigPDF: Qwen3 235B A22B Instruct 2507 on RTX 5060 report
RTX 3070Too bigPDF: Qwen3 235B A22B Instruct 2507 on RTX 3070 report
RTX 3080 10 GBToo bigPDF: Qwen3 235B A22B Instruct 2507 on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: Qwen3 235B A22B Instruct 2507 on RTX 3060 12 GB report
Arc B580Too bigPDF: Qwen3 235B A22B Instruct 2507 on Arc B580 report
RTX 4070Too bigPDF: Qwen3 235B A22B Instruct 2507 on RTX 4070 report
RTX 4070 SuperToo bigPDF: Qwen3 235B A22B Instruct 2507 on RTX 4070 Super report
RTX 5070Too bigPDF: Qwen3 235B A22B Instruct 2507 on RTX 5070 report
RTX 4060 Ti 16 GBToo bigPDF: Qwen3 235B A22B Instruct 2507 on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBToo bigPDF: Qwen3 235B A22B Instruct 2507 on RTX 5060 Ti 16 GB report
Arc A770 16 GBToo bigPDF: Qwen3 235B A22B Instruct 2507 on Arc A770 16 GB report
Radeon RX 9070 XTToo bigPDF: Qwen3 235B A22B Instruct 2507 on Radeon RX 9070 XT report
RTX 4070 Ti SuperToo bigPDF: Qwen3 235B A22B Instruct 2507 on RTX 4070 Ti Super report
RTX 4080 SuperToo bigPDF: Qwen3 235B A22B Instruct 2507 on RTX 4080 Super report
RTX 5070 TiToo bigPDF: Qwen3 235B A22B Instruct 2507 on RTX 5070 Ti report
RTX 5080Too bigPDF: Qwen3 235B A22B Instruct 2507 on RTX 5080 report
Radeon RX 7900 XTXToo bigPDF: Qwen3 235B A22B Instruct 2507 on Radeon RX 7900 XTX report
RTX 3090Too bigPDF: Qwen3 235B A22B Instruct 2507 on RTX 3090 report
RTX 4090Too bigPDF: Qwen3 235B A22B Instruct 2507 on RTX 4090 report
Radeon AI PRO R9700Too bigPDF: Qwen3 235B A22B Instruct 2507 on Radeon AI PRO R9700 report
RTX 5090Too bigPDF: Qwen3 235B A22B Instruct 2507 on RTX 5090 report
2× RTX 3090Too bigPDF: Qwen3 235B A22B Instruct 2507 on 2× RTX 3090 report
2× RTX 4090Too bigPDF: Qwen3 235B A22B Instruct 2507 on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GBToo bigPDF: Qwen3 235B A22B Instruct 2507 on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 AdaToo bigPDF: Qwen3 235B A22B Instruct 2507 on RTX 6000 Ada report
L40SToo bigPDF: Qwen3 235B A22B Instruct 2507 on L40S report
2× RTX 5090Too bigPDF: Qwen3 235B A22B Instruct 2507 on 2× RTX 5090 report
4× RTX 30902-bit32K55 tok/s1.3 sPDF: Qwen3 235B A22B Instruct 2507 on 4× RTX 3090 report
RTX PRO 6000 Blackwell2-bit44K82 tok/s1.2 sPDF: Qwen3 235B A22B Instruct 2507 on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: Qwen3 235B A22B Instruct 2507 on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: Qwen3 235B A22B Instruct 2507 on Mac mini M4 24 GB report
Mac mini M4 32 GBToo bigPDF: Qwen3 235B A22B Instruct 2507 on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: Qwen3 235B A22B Instruct 2507 on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GBToo bigPDF: Qwen3 235B A22B Instruct 2507 on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GBToo bigPDF: Qwen3 235B A22B Instruct 2507 on Mac mini M4 Pro 64 GB report
M4 Max 128 GB2-bit44K30 tok/s25 sPDF: Qwen3 235B A22B Instruct 2507 on M4 Max 128 GB report
M5 Max 128 GB2-bit44K32 tok/s7.1 sPDF: Qwen3 235B A22B Instruct 2507 on M5 Max 128 GB report
M3 Ultra 96 GBToo bigPDF: Qwen3 235B A22B Instruct 2507 on M3 Ultra 96 GB report
M3 Ultra 256 GB5-bit136K23 tok/s14 sPDF: Qwen3 235B A22B Instruct 2507 on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit256K18 tok/s14 sPDF: Qwen3 235B A22B Instruct 2507 on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit256K23 tok/s4.3 sPDF: Qwen3 235B A22B Instruct 2507 on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GBToo bigPDF: Qwen3 235B A22B Instruct 2507 on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB2-bit44K16 tok/s14 sPDF: Qwen3 235B A22B Instruct 2507 on Ryzen AI Max+ 395 128 GB report
RTX Spark3-bit32K18 tok/s5.0 sPDF: Qwen3 235B A22B Instruct 2507 on RTX Spark report
DGX Spark3-bit32K16 tok/s5.0 sPDF: Qwen3 235B A22B Instruct 2507 on DGX Spark report
Dual DGX Spark6-bit256K10.0 tok/s2.8 sPDF: Qwen3 235B A22B Instruct 2507 on Dual DGX Spark report
DGX Station GB3008-bit32K97 tok/s0.4 sPDF: Qwen3 235B A22B Instruct 2507 on DGX Station GB300 report
2× RTX PRO 6000 Blackwell5-bit136K55 tok/s0.7 sPDF: Qwen3 235B A22B Instruct 2507 on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit256K41 tok/s0.4 sPDF: Qwen3 235B A22B Instruct 2507 on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit256K41 tok/s0.2 sPDF: Qwen3 235B A22B Instruct 2507 on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXMToo bigPDF: Qwen3 235B A22B Instruct 2507 on A100 80 GB SXM report
H100 SXMToo bigPDF: Qwen3 235B A22B Instruct 2507 on H100 SXM report
4× H200 NVL8-bit256K79 tok/s0.4 sPDF: Qwen3 235B A22B Instruct 2507 on 4× H200 NVL report
8× H200 NVL8-bit256K79 tok/s0.2 sPDF: Qwen3 235B A22B Instruct 2507 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/Qwen3-235B-A22B-Instruct-2507-GGUF
vLLM
vllm serve Qwen/Qwen3-235B-A22B-Instruct-2507
MLX (Mac)
mlx_lm.chat --model mlx-community/Qwen3-235B-A22B-Instruct-2507-4bit

Questions

How much VRAM does Qwen3 235B A22B Instruct 2507 need?

Qwen3 235B A22B Instruct 2507 needs 133 GB for its weights at 4-bit, 135 GB with an 8K context and 139 GB with 32K. At 8-bit, 235 GB.

What is the minimum VRAM for Qwen3 235B A22B Instruct 2507?

About 84.4 GB at 2-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 235B A22B Instruct 2507?

235B parameters (22B active per token): 438 GB of weights at 16-bit, 133 GB at 4-bit.

Can Qwen3 235B A22B Instruct 2507 run on an RTX 4090?

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

More models