local/qwen-image-2512-20b Benchmark — 59/100 Overall, 192 Prompts, Every Image Shown

ImageBench V1 —

192 evaluations across 6 categories

Benchmark V1 verdicts are produced by VLM judges and can contain mistakes. Treat PASS/FAIL labels as machine-assisted assessments, and inspect the images yourself. Learn more about the methodology.

Generation Details

Source-backed model context, size, cost, and request settings for this run.

MakerAlibaba

FamilyQwen Image

Model size20B20B params

Costself-hostedfree

Latency80.2s median

Run targetlocal/qwen-image-2512-20b

Effective requestmodel: qwen-image-2512-20b · size: 1024x1024 · seed: 42

SourcesModel cardWeights & checkpoint

All 192 generations

Every challenge, grouped by category and subcategory. Solid-bordered tiles with a check passed the VLM judge; dashed tiles with a cross failed. Images are placeholders until generated.

Text Rendering

58%

Typography Style33%

Writing accuracy58%

Spatial Reasoning

43%

Attributes Binding33%

Compositionality33%

Counting56%

Negation33%

Relative Position25%

Scale & Proportions44%

Human realism

36%

Faces & Expressions33%

Full Body33%

Hands33%

Multi-Subject33%

Professional Studio

63%

Camera & Lighting42%

Color Precision67%

Photorealism33%

Graphical design

50%

Data Visualisation33%

Layout & Design44%

Style Diversity33%

Truthfulness

49%

Photorealism0%

Physics & Reflections33%

World Knowledge58%