VMetaphor-Bench (Descriptive Prompts): leaderboard
VMetaphor-Bench (Chen et al., 2026) with descriptive prompts that spell out how each of the 1,500 visual metaphors should look. A Qwen3.5-27B judge answers 9,594 multiple-choice fidelity questions about the generated images, and the board is the pooled accuracy in percent; higher is better.
Metric: MCQ accuracy (%; pooled over all 9,594 fidelity questions; 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; descriptive prompts that spell out the visual realization). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Nano Banana 2 (Gemini 3.1 Flash Image Preview) | 89.2 | |
| 2 | GPT Image 1.5 | 88.5 | |
| 3 | FLUX.2-dev | 87.3 | |
| 4 | Qwen-Image-2512 | 87 | |
| 5 | Seedream 5.0 Lite | 86.6 | |
| 6 | Z-Image-Turbo | 85.9 | |
| 7 | Janus-Pro-7B | 81 | #7341 |
| 8 | FLUX.1-dev | 80.7 | |
| 9 | BAGEL-7B-MoT | 79.1 | #5515 |
| 10 | Stable Diffusion 3.5 Large | 78.8 |
Interactive version: theaggregate.ai/benchmark?slug=vmetaphor-bench-descriptive-prompts · How It Works · Data refreshed daily, snapshot 2026-11-12.