Visual Aesthetic Benchmark (VAB): leaderboard
Multimodal large language models (MLLMs) are now routinely deployed for visual understanding, generation, and curation. A substantial fraction of these applications require an explicit aesthetic judgment. Most existing solutions reduce this judgment to predicting a scalar score for a single image. We first ask whether such scores faithfully capture comparative preference: in a controlled study wit
Metric: Overall Top-1 ap@1 (imported). Source: benchmarklist.com. Saturation forecast: Around May 2027. 23 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Median Expert | 77.7 | |
| 2 | Gemini 3 Pro | 55.4 | #80 |
| 3 | Gemini 3.1 Pro (Preview) | 54.8 | #58 |
| 4 | Claude Sonnet 4.6 | 54.8 | #90 |
| 5 | Claude Opus 4.6 | 53 | #63 |
| 6 | Gemini 3 Flash | 52.8 | #99 |
| 7 | GPT-5.1 | 52.3 | #135 |
| 8 | Seed 2.0 Pro | 51.3 | #118 |
| 9 | Claude Opus 4.5 | 51.2 | #83 |
| 10 | GPT-5 | 51 | #98 |
| 11 | Qwen 3.5 Plus | 50.4 | #125 |
| 12 | GPT-5.1 Codex | 50.2 | #159 |
| 13 | Qwen 3.5 397B A17B | 50.1 | #146 |
| 14 | Kimi K2.5 | 49.4 | #144 |
| 15 | GPT-4.1 | 49.3 | #248 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=visual-aesthetic-benchmark-vab · How It Works · Data refreshed daily, snapshot 2026-10-20.