ImageBench

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 ImageBench V1 run.

49.8
Overall
46%
Capability
53.2
Est. Preference
89
Pass
103
Fail
3.8s
Avg Latency
2.6s
Min Latency
19.5s
Max Latency
Text Rendering40%Spatial Reasoning54%Human realism24%Truthfulness26%Professional Studio89%Graphical design46%Preference53%Latency61%

All 192 generations

Text Rendering40%

Typography Style100%

Writing accuracy25%

Spatial Reasoning54%

Attributes Binding67%

Compositionality67%

Counting33%

Negation56%

Relative Position50%

Scale & Proportions56%

Human realism24%

Faces & Expressions50%

Full Body8%

Hands0%

Multi-Subject50%

Truthfulness26%

Photorealism0%

Physics & Reflections42%

World Knowledge17%

Professional Studio89%

Camera & Lighting83%

Color Precision100%

Photorealism67%

Graphical design46%

Data Visualisation0%

Layout & Design11%

Style Diversity83%