VMetaphor-Bench - Domain Presence: leaderboard
Domain presence in VMetaphor-Bench (Chen et al., 2026): two multiple-choice questions per generated image on whether the source and the target domain are recognizable, answered by a Qwen3.5-27B judge. Accuracy in percent; higher is better. Conceptual prompts, which state only the metaphorical idea.
Metric: MCQ accuracy (%; the two questions per image on whether both the source and the target domain appear; the generated image is judged by Qwen3.5-27B on multiple-choice questions (four options including cannot be determined) built from each metaphor's structured annotation; 1,500 visual metaphors curated from real creative imagery; conceptual prompts that state only the metaphorical idea). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT Image 1.5 | 93.6 | |
| 2 | Nano Banana 2 (Gemini 3.1 Flash Image Preview) | 93 | |
| 3 | Seedream 5.0 Lite | 90.6 | |
| 4 | FLUX.2-dev | 88.6 | |
| 5 | Qwen-Image-2512 | 87.3 | |
| 6 | Z-Image-Turbo | 84.3 | |
| 7 | Janus-Pro-7B | 83.4 | #7341 |
| 8 | Stable Diffusion 3.5 Large | 83 | |
| 9 | FLUX.1-dev | 82.3 | |
| 10 | BAGEL-7B-MoT | 79.8 | #5515 |
Interactive version: theaggregate.ai/benchmark?slug=vmetaphor-bench-domain-presence · How It Works · Data refreshed daily, snapshot 2026-11-12.