BizGenEval (Hard) - Knowledge-based Reasoning: leaderboard
Metric: Checklist score (0-100): for each generated image a Gemini 3 Flash Preview judge answers the prompt's 10 human-verified hard yes/no questions, the image scores max(0, 1 - 0.2 x errors), so five or more errors give 0, and scores are averaged and shown as a percentage, on the 100 knowledge-based reasoning prompts of BizGenEval (20 per document type: slides, webpages, posters, charts, scientific figures); one image per prompt at the official default settings, resolution set to the nearest supported aspect ratio of the reference design; higher is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 26 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Nano Banana Pro | 82.6 | |
| 2 | Nano Banana 2 | 64.6 | |
| 3 | Seedream 5.0 Lite | 41.8 | |
| 4 | GPT Image 1.5 | 26 | |
| 5 | Seedream 4.5 | 21.4 | |
| 6 | Wan2.6-T2I | 12.2 | |
| 7 | GPT Image 1 | 7.8 | |
| 8 | Seedream 4.0 | 6.8 | |
| 9 | EMU3.5 | 1.2 | |
| 10 | Z-Image | 0.6 |
Interactive version: theaggregate.ai/benchmark?slug=bizgeneval-hard-knowledge-based-reasoning · How It Works · Data refreshed daily, snapshot 2026-10-11.