Skip to content

Qwen3.5 122B A10B VRAM requirements

How much VRAM Qwen3.5 122B A10B needs
4-bit weights: 70.6 GBWith 8K context: 72.0 GBWith 32K context: 72.5 GB8-bit, 8K context: 125 GB
Parameters
125B
Active per token
10B
Longest context
256K
Released
Feb 2026
License
Apache 2.0
Strongest at
Chat, Coding, Agents, Docs

How much VRAM Qwen3.5 122B A10B needs

Memory Qwen3.5 122B A10B needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit233 GB234 GB235 GB237 GB
8-bit124 GB125 GB126 GB128 GB
6-bit95.5 GB96.9 GB97.4 GB99.7 GB
5-bit82.9 GB84.2 GB84.8 GB87.0 GB
4-bit70.6 GB72.0 GB72.5 GB74.8 GB
3-bit56.9 GB58.3 GB58.8 GB61.1 GB
2-bit43.7 GB45.0 GB45.6 GB47.8 GB
1.6-bit24.0 GB25.4 GB25.9 GB28.2 GB

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

GPUs and Macs that run Qwen3.5 122B A10B

The best way Qwen3.5 122B A10B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 4060Too bigPDF: Qwen3.5 122B A10B on RTX 4060 report
RTX 5060Too bigPDF: Qwen3.5 122B A10B on RTX 5060 report
RTX 3070Too bigPDF: Qwen3.5 122B A10B on RTX 3070 report
RTX 3080 10 GBToo bigPDF: Qwen3.5 122B A10B on RTX 3080 10 GB report
RTX 3060 12 GBToo bigPDF: Qwen3.5 122B A10B on RTX 3060 12 GB report
Arc B580Too bigPDF: Qwen3.5 122B A10B on Arc B580 report
RTX 4070Too bigPDF: Qwen3.5 122B A10B on RTX 4070 report
RTX 4070 SuperToo bigPDF: Qwen3.5 122B A10B on RTX 4070 Super report
RTX 5070Too bigPDF: Qwen3.5 122B A10B on RTX 5070 report
RTX 4060 Ti 16 GBToo bigPDF: Qwen3.5 122B A10B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GBToo bigPDF: Qwen3.5 122B A10B on RTX 5060 Ti 16 GB report
Arc A770 16 GBToo bigPDF: Qwen3.5 122B A10B on Arc A770 16 GB report
Radeon RX 9070 XTToo bigPDF: Qwen3.5 122B A10B on Radeon RX 9070 XT report
RTX 4070 Ti SuperToo bigPDF: Qwen3.5 122B A10B on RTX 4070 Ti Super report
RTX 4080 SuperToo bigPDF: Qwen3.5 122B A10B on RTX 4080 Super report
RTX 5070 TiToo bigPDF: Qwen3.5 122B A10B on RTX 5070 Ti report
RTX 5080Too bigPDF: Qwen3.5 122B A10B on RTX 5080 report
Radeon RX 7900 XTXToo bigPDF: Qwen3.5 122B A10B on Radeon RX 7900 XTX report
RTX 3090Too bigPDF: Qwen3.5 122B A10B on RTX 3090 report
RTX 4090Too bigPDF: Qwen3.5 122B A10B on RTX 4090 report
Radeon AI PRO R9700Too bigPDF: Qwen3.5 122B A10B on Radeon AI PRO R9700 report
RTX 5090Too bigPDF: Qwen3.5 122B A10B on RTX 5090 report
2× RTX 3090Too bigPDF: Qwen3.5 122B A10B on 2× RTX 3090 report
2× RTX 4090Too bigPDF: Qwen3.5 122B A10B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB2-bit32K142 tok/s0.9 sPDF: Qwen3.5 122B A10B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada2-bit32K115 tok/s0.7 sPDF: Qwen3.5 122B A10B on RTX 6000 Ada report
L40S2-bit32K107 tok/s1.1 sPDF: Qwen3.5 122B A10B on L40S report
2× RTX 50903-bit80K145 tok/s0.3 sPDF: Qwen3.5 122B A10B on 2× RTX 5090 report
4× RTX 30905-bit208K72 tok/s0.5 sPDF: Qwen3.5 122B A10B on 4× RTX 3090 report
RTX PRO 6000 Blackwell5-bit256K115 tok/s0.5 sPDF: Qwen3.5 122B A10B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GBToo bigPDF: Qwen3.5 122B A10B on Mac mini M4 16 GB report
Mac mini M4 24 GBToo bigPDF: Qwen3.5 122B A10B on Mac mini M4 24 GB report
Mac mini M4 32 GBToo bigPDF: Qwen3.5 122B A10B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GBToo bigPDF: Qwen3.5 122B A10B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GBToo bigPDF: Qwen3.5 122B A10B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB2-bit40K41 tok/s20 sPDF: Qwen3.5 122B A10B on Mac mini M4 Pro 64 GB report
M4 Max 128 GB5-bit256K41 tok/s10 sPDF: Qwen3.5 122B A10B on M4 Max 128 GB report
M5 Max 128 GB5-bit256K44 tok/s2.9 sPDF: Qwen3.5 122B A10B on M5 Max 128 GB report
M3 Ultra 96 GB3-bit256K62 tok/s5.6 sPDF: Qwen3.5 122B A10B on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit256K37 tok/s5.6 sPDF: Qwen3.5 122B A10B on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit256K37 tok/s5.6 sPDF: Qwen3.5 122B A10B on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit256K49 tok/s1.8 sPDF: Qwen3.5 122B A10B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB2-bit40K34 tok/s5.8 sPDF: Qwen3.5 122B A10B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB5-bit256K19 tok/s5.8 sPDF: Qwen3.5 122B A10B on Ryzen AI Max+ 395 128 GB report
RTX Spark6-bit256K23 tok/s2.0 sPDF: Qwen3.5 122B A10B on RTX Spark report
DGX Spark6-bit256K21 tok/s2.1 sPDF: Qwen3.5 122B A10B on DGX Spark report
Dual DGX Spark8-bit256K17 tok/s1.6 sPDF: Qwen3.5 122B A10B on Dual DGX Spark report
DGX Station GB3008-bit256K198 tok/s0.2 sPDF: Qwen3.5 122B A10B on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit256K87 tok/s0.3 sPDF: Qwen3.5 122B A10B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit256K87 tok/s0.1 sPDF: Qwen3.5 122B A10B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit256K87 tok/s0.1 sPDF: Qwen3.5 122B A10B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM4-bit176K138 tok/s1.3 sPDF: Qwen3.5 122B A10B on A100 80 GB SXM report
H100 SXM4-bit176K181 tok/s0.4 sPDF: Qwen3.5 122B A10B on H100 SXM report
4× H200 NVL8-bit256K164 tok/s0.1 sPDF: Qwen3.5 122B A10B on 4× H200 NVL report
8× H200 NVL8-bit256K164 tok/s0.1 sPDF: Qwen3.5 122B A10B 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

Ollama
ollama run qwen3.5:122b-a10b
llama.cpp
llama-server -hf unsloth/Qwen3.5-122B-A10B-GGUF
vLLM
vllm serve Qwen/Qwen3.5-122B-A10B
MLX (Mac)
mlx_lm.chat --model mlx-community/Qwen3.5-122B-A10B-4bit

Questions

How much VRAM does Qwen3.5 122B A10B need?

Qwen3.5 122B A10B needs 70.6 GB for its weights at 4-bit, 72.0 GB with an 8K context and 72.5 GB with 32K. At 8-bit, 125 GB.

What is the minimum VRAM for Qwen3.5 122B A10B?

About 45.0 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 RTX PRO 5000 Blackwell 48 GB.

How big is Qwen3.5 122B A10B?

125B parameters (10B active per token): 233 GB of weights at 16-bit, 70.6 GB at 4-bit.

Can Qwen3.5 122B A10B run on an RTX 4090?

Not on one card. The smallest machine here that runs it is the RTX PRO 5000 Blackwell 48 GB.

More models