Qwen-Image-Bench (English) - Real-World Fidelity: leaderboard
Metric: Real-world fidelity (knowledge-consistent depiction of real entities, text and physical facts) pillar Q-Judger score (0-100), the unified judge model fine-tuned from Qwen3.6-27B on expert annotations, aggregated bottom-up from 56 third-level facets, over the 1,000 creator-designed English prompts; higher is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 18 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT Image 2 | 59.4 |
| 2 | Nano Banana Pro | 55.91 |
| 3 | GPT Image 1.5 | 55.86 |
| 4 | Nano Banana 2 | 54.66 |
| 5 | Seedream 5.0 | 53.86 |
| 6 | Qwen Image 2.0 Pro | 53.06 |
| 7 | Seedream 4.0 | 52.68 |
| 8 | Seedream 4.5 | 51.55 |
| 9 | FLUX 2 Max | 50.69 |
| 10 | FLUX 2 Pro | 48.45 |
Interactive version: theaggregate.ai/benchmark?slug=qwen-image-bench-english-real-world-fidelity · How It Works · Data refreshed daily, snapshot 2026-10-07.