The AI Bench
Alibaba (Qwen) · Differs by model

Qwen-Image

Alibaba’s image models, strong at drawing readable text. 2512 is free for business; 2.1 is research-only.

How much memory it needs

Memory to run each model, from its model card or the size of its files. Smaller versions trade some quality for less memory.

ModelVersionNeedsBusiness use
Qwen-Image-2.14-bit GGUF11.2 GBNot allowed
BF1633.1 GB
Qwen-Image-25124-bit GGUF22.9 GBAllowed
BF1657.7 GB

Which computers run it

The best model in this group for each machine.

ComputerBest versionFit
Laptop or PC, no graphics card, 32 GBQwen-Image-2.1 4-bit GGUFNeeds a graphics card
RTX 4060 · 8 GBQwen-Image-2.1 4-bit GGUFSpills into system RAM
RTX 3060 · 12 GBQwen-Image-2.1 4-bit GGUFSpills into system RAM
RTX 5060 Ti · 16 GBQwen-Image-2.1 4-bit GGUFRuns well
RTX 4090 · 24 GBQwen-Image-2.1 4-bit GGUFRuns great
RTX 5090 · 32 GBQwen-Image-2512 4-bit GGUFRuns well
Mac with M4 · 16 GBQwen-Image-2.1 4-bit GGUFFits with a tweak
Mac with M4 Pro · 48 GBQwen-Image-2512 4-bit GGUFRuns great
Mac with M5 Max · 128 GBQwen-Image-2512 BF16Runs great
Ryzen AI Max+ 395 · 128 GBQwen-Image-2512 BF16Runs great
DGX Spark · 128 GBQwen-Image-2512 BF16Runs great
RTX PRO 6000 · 96 GBQwen-Image-2512 BF16Runs great
Mac Studio with M5 Ultra · 256 GBQwen-Image-2512 BF16Runs great
Mac Studio with M3 Ultra · 512 GBQwen-Image-2512 BF16Runs great

What it is good and bad at

Good at

  • Text inside images (signs, posters)
  • 2.1: transparent images and editing with references

Not good at

  • Large: 2512 needs about 23 GB even at 4-bit
  • 2.1 licence is research and evaluation only

Start it

  • Qwen-Image-2.1: runs in ComfyUI (also diffusers (QwenImage21Pipeline), stable-diffusion.cpp (GGUF)). Model card
  • Qwen-Image-2512: runs in ComfyUI (also diffusers, ComfyUI-GGUF, stable-diffusion.cpp (GGUF)). Model card