EditReward-Compass: leaderboard
Metric: Average preference accuracy (%) over all pairs 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 January 2027. 24 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 74.33 |
| 2 | Gemini 3 Flash | 72.68 |
| 3 | Qwen 3.6 27B | 71.83 |
| 4 | Qwen 3.5 35B A3B | 70.89 |
| 5 | Qwen 3.6 35B A3B | 70.51 |
| 6 | Qwen 3.5 27B | 69.98 |
| 7 | Gemma 4 31B | 67.09 |
| 8 | Qwen 3.5 27B (Non-reasoning) | 66.93 |
| 9 | Qwen 3.5 9B | 66.81 |
| 10 | GPT-4.1 | 66.11 |
| 11 | Qwen 3.6 27B (Non-reasoning) | 63.28 |
| 12 | Qwen 3.5 35B A3B (Non-reasoning) | 63.18 |
| 13 | Qwen 3.5 9B (Non-reasoning) | 60.16 |
| 14 | Qwen 3.6 35B A3B (Non-reasoning) | 59.95 |
| 15 | Gemma 4 26B A4B | 59.6 |
Interactive version: theaggregate.ai/benchmark?slug=editreward-compass · How It Works · Data refreshed daily, snapshot 2026-10-07.