# Test page

- Published: 2026-05-27
- Updated: 2026-06-10
- Author: Nikita Kornilov
- Canonical: https://dev.research.everypixel.com/test/

---

> **BLUF:** Enter your text

## Benchmark methodology
- **Prompts:** 35
- **Models:** GPT Image 2.0, NanoBanana 2, Midjourney v7
- **Seeds:** 0, 1, 2, 42
- **Dates:** Month 2026
- **Reviewer:** Sam E., Dennis S.

### Hit rate per model on e-commerce prompts

| Model | Hit rate | Cost / image | Latency (s) |
| --- | --- | --- | --- |
| GPT Image 2.0 | 0.87 | $0.005 | 8.2 |
| NanoBanana 2 | 0.71 | $0.004 | 5.4 |
| Midjourney v7 | 0.83 | $0.012 | 11.8 |
| Flux Pro 1.1 | 0.79 | $0.006 | 6.1 |

*Unit: 0..1 hit rate*

Source: https://research.everypixel.com/datasets/ecom-2026q2.csv

*Radar chart — Multi-dimensional comparison: GPT Image 2.0, NanoBanana 2, Midjourney v7 compared across Accuracy, Speed, Cost-eff, Composition, Detail, Brand-safety.*

Scores normalized to 0–1; higher is better.

## Model: [GPT Image 2.0](/model-cards/gpt-image-2/)
- **Provider:** OpenAI
- **License:** Proprietary
- **Cost:** $0.005 / image
- **Hit rate:** 0.87
- **Best for:** Product photography, brand-safe outputs
- **Fails on:** Stylized illustration, hands at edge of frame

**Prompt Pack:** [E-commerce prompts pack](https://research.everypixel.com/datasets/ecom-prompts-2026q2.csv) — CSV, 35 prompts, CC-BY 4.0

## Frequently asked questions
### Question

Answer

**Raw data:** [Raw experiment data](https://research.everypixel.com/datasets/ecom-2026q2-raw.csv) — CSV, CC-BY 4.0, [schema](https://research.everypixel.com/schemas/benchmark-v1.json)

## Changelog
- **2026-06-04** — Перезамер на GPT Image 2.0; обновлена методология.
- **2026-05-12** — Первая публикация бенчмарка.