EditReward-Compass - Visual Quality: leaderboard
Metric: Preference accuracy (%) on the pairs that differ in visual quality on EditReward-Compass (2,251 human-verified preference pairs of edited images sampled from the same editing models under the same instruction), general MLLMs prompted with the Edit-Compass rubric; thinking disabled unless the row is a thinking run; higher is better. Source: arxiv.org. Saturation forecast: Around August 2028. 24 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 3.5 27B | 58.78 |
| 2 | Qwen 3.6 27B | 57.43 |
| 3 | Qwen 3.5 35B A3B | 56.08 |
| 4 | Qwen 3.6 35B A3B | 56.08 |
| 5 | Qwen 3.5 27B (Non-reasoning) | 53.81 |
| 6 | GPT-4.1 | 53.38 |
| 7 | Gemma 4 31B | 49.32 |
| 8 | Qwen 3.5 9B | 48.98 |
| 9 | Gemini 3 Flash | 48.65 |
| 10 | Qwen 3.5 9B (Non-reasoning) | 46.35 |
| 11 | Qwen 3.5 2B | 44.66 |
| 12 | Gemini 3.1 Pro (Preview) | 44.59 |
| 13 | Gemma 4 26B A4B | 43.92 |
| 14 | Qwen 3.6 35B A3B (Non-reasoning) | 43.44 |
| 15 | Qwen 3.5 35B A3B (Non-reasoning) | 42.05 |
Interactive version: theaggregate.ai/benchmark?slug=editreward-compass-visual-quality · How It Works · Data refreshed daily, snapshot 2026-10-07.