whatmodel.sh / image / Qwen-Image

Qwen-Image

+ contribute
Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

Qwen-Image is an open-weights image model from Alibaba, released August 2025. Its published size is 20B parameters. Licence: apache-2.0.

54.4±1.3#69 of 144 image modelsfrom 14 benchmark results across 14 sources

Score within the LMArena text-to-image comparison set only.

$ how it compares

The image models ranked either side of Qwen-Image, out of 197 with a score, then more from Alibaba.

#modelscorewhereRAMreleased
83FLUX.1 Kontext [max] Black Forest Labs55±2api $0.080/image2025-05
84Reve Edit Fast Reve55±4comparison only2025-11
85Qwen-Image Alibaba54±1api $0.025/image +local~13GB2025-08
86HiDream-O1-Image-Dev HiDream54±5comparison only2026-05
87LLaDA-Image-Turbo Ant Group54±3local~22GB2026-09
more from Alibaba
11Qwen-Image-3.0-Pro Alibaba75±2api $0.040/image2026-07
18Qwen-Image-2.1 Alibaba69±2local~12GB2026-09
20Qwen-Image-2.0-Pro (2026-06-22) Alibaba67±2api $0.075/image2026-06

see all image models · find the ones that fit your machine

$ what it's good at

text to image▂▅▅▃▄▄█▃▁▂▁▁ #62 of 126 · 88% of its score · 16 tests

Trails LongCat-Image, LLaDA-Image-Turbo here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
LMArena text-to-image1056.8 ✎▂▂▃▄▇█▄▂▄▂·▂ beats 16/75avg 1027▇▇▇▇▇▇▇▇▇▇ 25%typical
T2I-ReasonBench (overall)67.5% ✎▅▂█▂▂▃····▂▂ beats 11/13easy · avg 69%▇▇▇▇ 11%typical
ServImage task acceptance (human payment decisions, commercial design tasks)24.9% ✎▂▃··▂··▃█··▂ beats 4/11hard · avg 18%▇▇▇▇ 9.1%sharp
QC-T2I-Bench (English, HCQ total)87.7% ✎▄····▄█▄▄█▄█ beats 7/10easy · avg 82%▇▇▇▇ 8.9%typical
GenExam (relaxed overall, GPT-5 judge)27.5% ✎▆▆·▂▂····▂▂█ beats 4/13hard · avg 20%▇▇▇ 8.3%sharp · at its level · noisy
TIIF-Bench v3 (short prompts, TIIF-Evaluator)87.0% ✎▂█····▄▄▄··▂ beats 7/11easy · avg 79%▇▇▇ 8.2%typical
UltraText Bench (Composite, bilingual dense text rendering)41.3% ✎▁▂▂█▃▁▂▃·▁·▁ beats 8/22hard · avg 28%▇▇▇ 6.5%sharp · at its level · noisy
ImagenWorld text-guided generation (human overall)87.0% ✎▄··█▄·▄··▄·▄ beats 5/6easy · avg 84%▇▇ 5.5%typical
8 more tests
testresultvs the others in its rundifficultycountswhy
WISE (WiScore overall, GPT-4o judge)62.0% ✎▄·▆▂▆█▄▆▂▄·▄ beats 19/22medium · avg 60%▇ 3.1%at its level
Qwen-Image-Bench (Q-Judger, v1) lab49.2% ✎█▅▅·▅▅█·█··▃ beats 2/17medium · avg 46%▇ 2.7%blunt · at its level
ChineseWord (LongCat-Image 8,105-character set, PPOCRv5) overall56.6% ✎▄█·········▄ beats 1/3medium · avg 58%▇ 0.5%blunt · at its level · noisy
DPG-Bench lab88.3% ✎▄·▂·▂▆█▄▂▄▂▂ beats 15/17—not counted yet: no other model in the same run to compare it with
LongText-Bench (EN) lab94.3% ✎▃·····▅██▅▃▃ beats 7/12—not counted yet: no other model in the same run to compare it with
LongText-Bench (CN) lab94.6% ✎▃·······████ beats 8/12—not counted yet: no other model in the same run to compare it with
OneIG-Bench (EN) lab53.9% ✎▃······▃▅█▅▃ beats 6/9—not counted yet: no other model in the same run to compare it with
OneIG-Bench (CN) lab54.8% ✎▃·▃█▃·▃··▃·▃ beats 8/8—not counted yet: no other model in the same run to compare it with
image to image / editing▁▁·▃▂▃█▃▃▄▁▁ #46 of 62 * · 9.1% of its score · 1 test

Beats GPT Image 1 Mini, FLUX.1 Kontext [max] here, though they rank higher overall. Trails GLM-Image, FLUX.2 [klein] 4B, Ideogram 3.0 here, though they rank lower overall.

testresultvs the others in its rundifficultycountswhy
ServImage task acceptance (human payment decisions, commercial design tasks)24.9% ✎▂▃··▂··▃█··▂ beats 4/11hard · avg 18%▇▇▇▇ 9.1%sharp
other tests3 not in a task · 11% of its score
testresultvs the others in its rundifficultycountswhy
Rapidata human-face generation alignment (Elo)1110.5 ✎▃··▅▃··▃▃··█ beats 7/8avg 1012▇▇▇ 6.3%typical
Rapidata Benchmark.ai image benchmark (overall Elo)991.7 ✎▁▁▁▄▂█▂▁▁▂▁▁ beats 15/34avg 875▇▇ 3.8%typical
Knowledge Visualization T2I (Brief Caption, English overall)25.7% ✎▅·▃·▃··▃·█▃▃ beats 3/9hard · avg 28%▇ 0.8%blunt · at its level

$ where to run it

Through an API

api · Replicate · $0.025/image · replicate.com ↗ · price sourceUSD per output image on Replicate's qwen/qwen-image model (billing: per output image; hardware time not added). The page's current version is served from replicate-pruna/qwen-image (Pruna-optimized deployment). Product: Replicate model API. API model: qwen/qwen-image. Settings: one output image. Price basis: published list rate. Currency: USD. Replicate official model page, observed 2026-09-25.

Other ways to run it

version

  1. Run it as published · needs ~12.5 GB, NVIDIA only · from huggingface.co/docs
    uv run --with diffusers --with transformers --with accelerate --with torch --with sentencepiece --with bitsandbytes python -c "import torch; from diffusers import DiffusionPipeline; from diffusers.quantizers import PipelineQuantizationConfig; q=PipelineQuantizationConfig(quant_backend='bitsandbytes_4bit', quant_kwargs={'load_in_4bit':True,'bnb_4bit_quant_type':'nf4','bnb_4bit_compute_dtype':torch.bfloat16}, components_to_quantize=['transformer','text_encoder']); p=DiffusionPipeline.from_pretrained('Qwen/Qwen-Image', torch_dtype=torch.bfloat16, quantization_config=q, device_map='cuda'); p('a cat holding a sign', negative_prompt=' ', num_inference_steps=50, true_cfg_scale=4.0).images[0].save('out.png')"

Official weights and documentation: huggingface.co/Qwen ↗

$ how the score was worked out · LMArena leaderboard dataset 25%, T2I-ReasonBench: Benchmarking… 11%, ServImage: An Image Generatio… 9% and 11 more

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

sourceresultsshare of the scorelicence
LMArena leaderboard dataset1▇▇▇▇▇▇▇ 25%CC BY 4.0
T2I-ReasonBench: Benchmarking Reasoning-Informed Text-to-Image Generation (Findings of ACL 2026)1▇▇▇ 11%CC-BY-4.0
ServImage: An Image Generation and Editing Benchmark from Real-world Commercial Imaging Services (arXiv:2604.24023v2)1▇▇▇ 9%CC-BY-4.0
QC-T2I-Bench: Scalable Question-Centric Text-to-Image Evaluation (Reliable Ranking, Fine-Grained Diagnosis, and Cost-Aware Routing)1▇▇▇ 9%CC-BY-4.0
GenExam leaderboard (OpenGVLab/GenExam README, relaxed score)1▇▇ 8%MIT
TIIF-Bench: How Does Your T2I Model Follow Your Instructions? (arXiv v3)1▇▇ 8%CC-BY-4.0
UltraText Bench: A Comprehensive Bilingual Benchmark for Evaluating Visual Text Rendering in Image Generation (arXiv:2610.09823v1)1▇▇ 7%CC-BY-4.0
Rapidata/Face_Generation_Benchmark (Human Face Generation Alignment) published dataset-card leaderboard1▇▇ 6%CDLA-Permissive-2.0
ImagenWorld: Stress-Testing Image Generation Models with Explainable Human Evaluation on Open-ended Real-World Tasks1▇▇ 5%CC-BY-4.0
Rapidata/text-to-image-human-preference-6M (Benchmark.ai Image Benchmark) published dataset-card rankings1▇ 4%CC-BY-4.0
WISE: A World Knowledge-Informed Semantic Evaluation for Text-to-Image Generation (arXiv v4)1▇ 3%CC-BY-SA-4.0
Qwen-Image-Bench: From Generation to Creation in Text-to-Image Evaluation · lab-reported1▇ 3%CC-BY-4.0
Knowledge Visualization: A Benchmark and Method for Knowledge-Intensive Text-to-Image Generation (arXiv:2604.22302v1)1▇ 1%CC-BY-4.0
Meituan LongCat — LongCat-Image technical report1▇ 0%CC BY 4.0
prior · pull toward the middle▇ 1%
weighted average= 54.4

not counted (5)

Qwen/Qwen-Image model cardDPG-Bench [correct]88.3%fewer than three models have this test yet, or this source ran it for this model alone
Qwen/Qwen-Image model cardLongText-Bench (CN) [correct]94.6%fewer than three models have this test yet, or this source ran it for this model alone
Qwen/Qwen-Image model cardLongText-Bench (EN) [correct]94.3%fewer than three models have this test yet, or this source ran it for this model alone
Qwen/Qwen-Image model cardOneIG-Bench (CN) [correct]54.8%fewer than three models have this test yet, or this source ran it for this model alone
Qwen/Qwen-Image model cardOneIG-Bench (EN) [correct]53.9%fewer than three models have this test yet, or this source ran it for this model alone

$ improve this page

Update model informationAdd a place to run itAdd a test resultPropose a benchmarkCorrect a benchmark scoreEdit data on GitHub

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