DistortBench: leaderboard
Metric: Joint accuracy (%) on DistortBench's 13,500 four-choice questions (27 distortion types at five KADID-10k calibrated severity levels on 100 KADIS-700k photos, no reference image), the answer naming both type and severity, answered-only accuracy (unparseable responses, at most 3.4 percent, leave the denominator), chance 25; higher is better. Source: arxiv.org. Saturation forecast: Around January 2028. 18 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 3.5 27B (Non-reasoning) | 61.9 |
| 2 | Qwen 3.5 27B | 58.2 |
| 3 | Qwen 3.5 35B A3B (Non-reasoning) | 56.8 |
| 4 | Qwen 3.5 4B (Non-reasoning) | 56.1 |
| 5 | Qwen 3.5 9B | 55 |
| 6 | Qwen 3.5 35B A3B | 54.6 |
| 7 | Qwen 3.5 4B | 53.1 |
| 8 | Qwen 3.5 9B (Non-reasoning) | 51.8 |
| 9 | GPT-5.4 | 51.6 |
| 10 | Gemma 3 27B | 47.5 |
| 11 | Gemma 3 12B | 46.6 |
| 12 | Qwen 2 VL 7B | 36.7 |
| 13 | Gemma 3 4B | 33.7 |
Interactive version: theaggregate.ai/benchmark?slug=distortbench · How It Works · Data refreshed daily, snapshot 2026-10-07.