VMetaphor-Bench - Metaphoric Efficacy: leaderboard
Metaphoric efficacy in VMetaphor-Bench (Chen et al., 2026): a Qwen3.5-27B judge rates from 1 to 5 how clearly each generated image communicates the intended metaphor without textual explanation. Mean rating; higher is better. Conceptual prompts, which state only the metaphorical idea.
Metric: Judge rating (1-5) (Qwen3.5-27B rates each generated image from 1 to 5 against the metaphor's structured annotation, 1,500 visual metaphors; this board is the dimension metaphoric efficacy: how clearly the image conveys the intended metaphorical meaning without text; 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 | 4.27 | |
| 2 | Nano Banana 2 (Gemini 3.1 Flash Image Preview) | 4.05 | |
| 3 | Seedream 5.0 Lite | 3.78 | |
| 4 | FLUX.2-dev | 3.48 | |
| 5 | Qwen-Image-2512 | 3.39 | |
| 6 | Stable Diffusion 3.5 Large | 2.99 | |
| 7 | Z-Image-Turbo | 2.93 | |
| 8 | FLUX.1-dev | 2.9 | |
| 9 | Janus-Pro-7B | 2.9 | #7341 |
| 10 | BAGEL-7B-MoT | 2.69 | #5515 |
Interactive version: theaggregate.ai/benchmark?slug=vmetaphor-bench-metaphoric-efficacy · How It Works · Data refreshed daily, snapshot 2026-11-12.