Skip to content

Qwen3.5 0.8B VRAM requirements

How much VRAM Qwen3.5 0.8B needs
4-bit weights: 505 MBWith 8K context: 1.6 GBWith 32K context: 1.8 GB8-bit, 8K context: 1.9 GB
Parameters
873M
Longest context
256K
Released
Feb 2026
License
Apache 2.0
Strongest at
Text and images

How much VRAM Qwen3.5 0.8B needs

Memory Qwen3.5 0.8B needs at each precision and context length
PrecisionWeightsWith 8KWith 32KWith 128K
16-bit1.6 GB2.7 GB3.0 GB4.1 GB
8-bit885 MB1.9 GB2.2 GB3.3 GB
6-bit683 MB1.7 GB2.0 GB3.1 GB
5-bit592 MB1.6 GB1.9 GB3.1 GB
4-bit505 MB1.6 GB1.8 GB3.0 GB
3-bit407 MB1.5 GB1.7 GB2.9 GB
2-bit312 MB1.4 GB1.7 GB2.8 GB
1.6-bit172 MB1.2 GB1.5 GB2.6 GB

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

GPUs and Macs that run Qwen3.5 0.8B

The best way Qwen3.5 0.8B runs on common GPUs, Macs and AI boxes
MachineBest fitContextSpeedFirst replyReport
RTX 40608-bit256K176 tok/s0.1 sPDF: Qwen3.5 0.8B on RTX 4060 report
RTX 50608-bit256K267 tok/s0.1 sPDF: Qwen3.5 0.8B on RTX 5060 report
RTX 30708-bit256K267 tok/s0.1 sPDF: Qwen3.5 0.8B on RTX 3070 report
RTX 3080 10 GB8-bit256K400 tok/s0.1 sPDF: Qwen3.5 0.8B on RTX 3080 10 GB report
RTX 3060 12 GB8-bit256K223 tok/s0.1 sPDF: Qwen3.5 0.8B on RTX 3060 12 GB report
Arc B5808-bit256K179 tok/s0.1 sPDF: Qwen3.5 0.8B on Arc B580 report
RTX 40708-bit256K294 tok/s0.1 sPDF: Qwen3.5 0.8B on RTX 4070 report
RTX 4070 Super8-bit256K294 tok/s0.1 sPDF: Qwen3.5 0.8B on RTX 4070 Super report
RTX 50708-bit256K366 tok/s0.1 sPDF: Qwen3.5 0.8B on RTX 5070 report
RTX 4060 Ti 16 GB8-bit256K184 tok/s0.1 sPDF: Qwen3.5 0.8B on RTX 4060 Ti 16 GB report
RTX 5060 Ti 16 GB8-bit256K267 tok/s0.1 sPDF: Qwen3.5 0.8B on RTX 5060 Ti 16 GB report
Arc A770 16 GB8-bit256K207 tok/s0.1 sPDF: Qwen3.5 0.8B on Arc A770 16 GB report
Radeon RX 9070 XT8-bit256K273 tok/s0.1 sPDF: Qwen3.5 0.8B on Radeon RX 9070 XT report
RTX 4070 Ti Super8-bit256K366 tok/s0.1 sPDF: Qwen3.5 0.8B on RTX 4070 Ti Super report
RTX 4080 Super8-bit256K391 tok/s0.1 sPDF: Qwen3.5 0.8B on RTX 4080 Super report
RTX 5070 Ti8-bit256K448 tok/s0.1 sPDF: Qwen3.5 0.8B on RTX 5070 Ti report
RTX 50808-bit256K469 tok/s0.1 sPDF: Qwen3.5 0.8B on RTX 5080 report
Radeon RX 7900 XTX8-bit256K351 tok/s0.1 sPDF: Qwen3.5 0.8B on Radeon RX 7900 XTX report
RTX 30908-bit256K462 tok/s0.1 sPDF: Qwen3.5 0.8B on RTX 3090 report
RTX 40908-bit256K485 tok/s0.1 sPDF: Qwen3.5 0.8B on RTX 4090 report
Radeon AI PRO R97008-bit256K273 tok/s0.1 sPDF: Qwen3.5 0.8B on Radeon AI PRO R9700 report
RTX 50908-bit256K678 tok/s0.1 sPDF: Qwen3.5 0.8B on RTX 5090 report
2× RTX 30908-bit256K462 tok/s0.1 sPDF: Qwen3.5 0.8B on 2× RTX 3090 report
2× RTX 40908-bit256K485 tok/s0.1 sPDF: Qwen3.5 0.8B on 2× RTX 4090 report
RTX PRO 5000 Blackwell 48 GB8-bit256K579 tok/s0.1 sPDF: Qwen3.5 0.8B on RTX PRO 5000 Blackwell 48 GB report
RTX 6000 Ada8-bit256K469 tok/s0.1 sPDF: Qwen3.5 0.8B on RTX 6000 Ada report
L40S8-bit256K437 tok/s0.1 sPDF: Qwen3.5 0.8B on L40S report
2× RTX 50908-bit256K678 tok/s0.1 sPDF: Qwen3.5 0.8B on 2× RTX 5090 report
4× RTX 30908-bit256K462 tok/s0.1 sPDF: Qwen3.5 0.8B on 4× RTX 3090 report
RTX PRO 6000 Blackwell8-bit256K678 tok/s0.1 sPDF: Qwen3.5 0.8B on RTX PRO 6000 Blackwell report
Mac mini M4 16 GB8-bit256K87 tok/s0.9 sPDF: Qwen3.5 0.8B on Mac mini M4 16 GB report
Mac mini M4 24 GB8-bit256K87 tok/s0.9 sPDF: Qwen3.5 0.8B on Mac mini M4 24 GB report
Mac mini M4 32 GB8-bit256K87 tok/s0.9 sPDF: Qwen3.5 0.8B on Mac mini M4 32 GB report
Mac mini M4 Pro 24 GB8-bit256K163 tok/s0.5 sPDF: Qwen3.5 0.8B on Mac mini M4 Pro 24 GB report
Mac mini M4 Pro 48 GB8-bit256K163 tok/s0.5 sPDF: Qwen3.5 0.8B on Mac mini M4 Pro 48 GB report
Mac mini M4 Pro 64 GB8-bit256K163 tok/s0.5 sPDF: Qwen3.5 0.8B on Mac mini M4 Pro 64 GB report
M4 Max 128 GB8-bit256K236 tok/s0.2 sPDF: Qwen3.5 0.8B on M4 Max 128 GB report
M5 Max 128 GB8-bit256K251 tok/s0.1 sPDF: Qwen3.5 0.8B on M5 Max 128 GB report
M3 Ultra 96 GB8-bit256K270 tok/s0.1 sPDF: Qwen3.5 0.8B on M3 Ultra 96 GB report
M3 Ultra 256 GB8-bit256K270 tok/s0.1 sPDF: Qwen3.5 0.8B on M3 Ultra 256 GB report
M3 Ultra 512 GB8-bit256K270 tok/s0.1 sPDF: Qwen3.5 0.8B on M3 Ultra 512 GB report
M5 Ultra 512 GB8-bit256K319 tok/s0.1 sPDF: Qwen3.5 0.8B on M5 Ultra 512 GB report
Ryzen AI Max+ 395 64 GB8-bit256K136 tok/s0.1 sPDF: Qwen3.5 0.8B on Ryzen AI Max+ 395 64 GB report
Ryzen AI Max+ 395 128 GB8-bit256K136 tok/s0.1 sPDF: Qwen3.5 0.8B on Ryzen AI Max+ 395 128 GB report
RTX Spark8-bit256K191 tok/s0.1 sPDF: Qwen3.5 0.8B on RTX Spark report
DGX Spark8-bit256K176 tok/s0.1 sPDF: Qwen3.5 0.8B on DGX Spark report
Dual DGX Spark8-bit256K176 tok/s0.1 sPDF: Qwen3.5 0.8B on Dual DGX Spark report
DGX Station GB3008-bit256K1,098 tok/s0.1 sPDF: Qwen3.5 0.8B on DGX Station GB300 report
2× RTX PRO 6000 Blackwell8-bit256K678 tok/s0.1 sPDF: Qwen3.5 0.8B on 2× RTX PRO 6000 Blackwell report
4× RTX PRO 6000 Blackwell8-bit256K678 tok/s0.1 sPDF: Qwen3.5 0.8B on 4× RTX PRO 6000 Blackwell report
8× RTX PRO 6000 Blackwell8-bit256K678 tok/s0.1 sPDF: Qwen3.5 0.8B on 8× RTX PRO 6000 Blackwell report
A100 80 GB SXM8-bit256K723 tok/s0.1 sPDF: Qwen3.5 0.8B on A100 80 GB SXM report
H100 SXM8-bit256K890 tok/s0.1 sPDF: Qwen3.5 0.8B on H100 SXM report
4× H200 NVL8-bit256K998 tok/s0.1 sPDF: Qwen3.5 0.8B on 4× H200 NVL report
8× H200 NVL8-bit256K998 tok/s0.1 sPDF: Qwen3.5 0.8B 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 turintech/Qwen3.5-0.8B-Artemis-GGUF
vLLM
vllm serve Qwen/Qwen3.5-0.8B
MLX (Mac)
mlx_lm.chat --model mlx-community/Qwen3.5-0.8B-MLX-4bit

Questions

How much VRAM does Qwen3.5 0.8B need?

Qwen3.5 0.8B needs 505 MB for its weights at 4-bit, 1.6 GB with an 8K context and 1.8 GB with 32K. At 8-bit, 1.9 GB.

What is the minimum VRAM for Qwen3.5 0.8B?

About 1.5 GB at 3-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 4060.

How big is Qwen3.5 0.8B?

873M parameters: 1.6 GB of weights at 16-bit, 505 MB at 4-bit.

Can Qwen3.5 0.8B run on an RTX 4090?

Yes. At 8-bit it holds 256K of context and writes about 485 tokens a second.

More models