A side-by-side benchmark comparison of the strongest ImageBench V1 models, including overall pass rate, category strengths, and generated-image evidence.
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.
| Model | Overall | Pass / Fail | Avg latency | Benchmark | Text Rendering | Spatial Reasoning | Human realism | Truthfulness | Professional Studio | Graphical design |
|---|---|---|---|---|---|---|---|---|---|---|
| openai/gpt-image-2 | 83.9% | 161 / 31 | 45.3s | View result | 100.0% | 91.2% | 54.8% | 74.1% | 100.0% | 100.0% |
| fal/google/nano-banana-2 | 73.4% | 141 / 51 | 28.1s | View result | 73.3% | 87.7% | 38.1% | 55.6% | 92.6% | 100.0% |
| bfl/flux-2-max | 62.5% | 120 / 72 | 26.7s | View result | 93.3% | 75.4% | 23.8% | 40.7% | 88.9% | 75.0% |
| bfl/flux-2-pro | 63.0% | 121 / 71 | 11.8s | View result | 80.0% | 75.4% | 31.0% | 44.4% | 85.2% | 75.0% |
Open the full ImageBench V1 result page for each model, including category breakdowns, prompts, generated images, and pass/fail judgments.
openai/gpt-image-2 leads this category at 100.0% (15/15 pass).
A poster with headline 'SUMMER SALE' in large bold text, body text 'Up to 50% off all items this weekend only' in medium size, and fine print 'Terms and conditions apply. See store for details.' at the bottom in small text
openai/gpt-image-2 leads this category at 91.2% (52/57 pass).
A pink fire truck, a blue banana, and a green sun in a surreal landscape
openai/gpt-image-2 leads this category at 54.8% (23/42 pass).
An elderly woman in her 80s with deep wrinkles and white hair, smiling warmly with kind eyes
openai/gpt-image-2 leads this category at 74.1% (20/27 pass).
A row of trees reflected in a calm lake at sunset, with the reflection slightly darker and gently rippled
openai/gpt-image-2 leads this category at 100.0% (27/27 pass).
A portrait shot at 85mm f/1.4 with creamy bokeh background, backlit with warm rim light separating the subject from a blurred garden
openai/gpt-image-2 leads this category at 100.0% (24/24 pass).
A magazine double-page spread with a left page containing a large hero photo, and a right page with a two-column article, a pull quote, a sidebar with author bio, and a page number