whatmodel.sh / image / Qwen-Image-2.1
Qwen-Image-2.1
↗ share
+ contribute
Qwen-Image-2.1 is an open-weights image model from Alibaba, released September 2026. Its published size is 7B parameters. Licence: qwen-research.
Catalogue identity notes
Comparison-only. Official repo Qwen/Qwen-Image-2.1 @790c926 (created 2026-09-14), Qwen Research License Agreement; parameter count was absent from that original card revision; see params_b_source for the newer evidence. Arena entry 'qwen-image-2.1' (alibaba, qwen-research), first listed 2026-09-22. Not Qwen-Image 2.0 Pro.
Model information sources
Release date ↗ · Parameter count ↗
Covers the visual generation component (32-layer single-stream DiT) only, not the text encoder/VAE. The card at the revision cited in identity_note (790c926) did not state it; the current card (d26bb61) does. The official GitHub README also says: "Transformer: 32 layers, 7B parameters".
Score within the LMArena text-to-image comparison set only.
$ how it compares
The image models ranked either side of Qwen-Image-2.1, out of 197 with a score, then more from Alibaba.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 16 |  Nano Banana 2 Lite (Gemini 3.1 Flash-Lite Image) | 73±2 | api $0.034/image | ||
| 17 |  GPT Image 1.5 | 71±1 | api $0.034/image | ||
| 18 | Qwen-Image-2.1 | 69±2 | local | ||
| 19 |  Ideogram 4.5 | 68±3 | comparison only | ||
| 20 | Qwen-Image-2.0-Pro (2026-06-22) | 67±2 | api $0.075/image | ||
| more from Alibaba | |||||
| 11 | Qwen-Image-3.0-Pro | 75±2 | api $0.040/image | ||
| 24 | Qwen-Image 2.0 Pro | 66±3 | api $0.075/image | ||
| 33 |  Wan 2.7 Image Pro | 63±2 | api $0.075/image | ||
$ what it's good at
text to image▂▅▅▃▄▄█▃▁▂▁▁ #17 of 126 * · 77% of its score · 2 tests
Trails Qwen-Image 2.0 Pro, Ideogram 4.0 here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| LMArena text-to-image | 1222.9 ✎ | ▂▂▃▄▇█▄▃▄▂·▂ beats 59/75 | 57% |
| GenExam (relaxed overall, GPT-5 judge) | 47.1% ✎ | ▆▆·▃▂····▂▂█ beats 6/13 | 20% |
image to image / editing▁▁·▃▂▃█▃▃▄▁▁ #13 of 62 * · 22% of its score · 1 test
Beats Reve 2.0, Nano Banana 2 Lite (Gemini 3.1 Flash-Lite Image), GPT Image 1.5 here, though they rank higher overall. Trails Grok Imagine Image Pro here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Arena image edit | 1368.7 ✎ | ▂▁▁▂▆▇█▄▆▅▁▂ beats 39/53 | 22% |
$ where to run it
Other ways to run it
version
- Run it as published · needs ~16.1 GB · from github.com/leejet
sd-cli --diffusion-model models/qwen_image_2.1_int8_convrot.safetensors --vae models/qwen_image_2.1_vae_bf16.safetensors --llm models/Qwen3VL-8B-Instruct-Q4_K_M.gguf -p "a lovely cat holding a sign says 'qwen2.1.cpp'" --cfg-scale 6.0 --sampling-method euler -v --offload-to-cpu --fa -o qwen_image_2.1.png
Official weights and documentation: huggingface.co/Qwen ↗
$ how the score was worked out · LMArena leaderboard dataset 78%, GenExam leaderboard (OpenGVLa… 20%
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 LMArena leaderboard dataset | 2 | 78% |
 GenExam leaderboard (OpenGVLab/GenExam README, relaxed score) | 1 | 20% |
| prior · pull toward the middle | 2% | |
| weighted average | = 68.8 |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json