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.

local/bonsai-image-ternary-4b

Local

Ternary Bonsai Image 4B is PrismML's compact ternary-weight image model, derived for local deployment with a much smaller diffusion-transformer footprint while retaining much of FLUX.2 Klein 4B's practical behavior.

Maker
PrismML
Family
Bonsai Image 4B
Model Size
4B
high
Cost
local run; no API price
not_applicable
Run Target
gx10/bonsai-image-4b
Effective Request
Effective request fields unknown
Sources
PrismML Bonsai Image 4B announcementBonsai Image ternary model card
46.5
Overall
43%
Capability
49.9
Est. Preference
83
Pass
109
Fail
4.1s
Avg Latency
3.4s
Min Latency
94.2s
Max Latency
Text Rendering27%Spatial Reasoning60%Human realism17%Truthfulness30%Professional Studio78%Graphical design38%Preference50%Latency59%

All 192 generations

Text Rendering27%

Typography Style33%

Writing accuracy25%

Spatial Reasoning60%

Attributes Binding100%

Compositionality67%

Counting33%

Negation33%

Relative Position75%

Scale & Proportions44%

Human realism17%

Faces & Expressions50%

Full Body0%

Hands8%

Multi-Subject0%

Truthfulness30%

Photorealism0%

Physics & Reflections42%

World Knowledge25%

Professional Studio78%

Camera & Lighting83%

Color Precision75%

Photorealism67%

Graphical design38%

Data Visualisation0%

Layout & Design0%

Style Diversity75%