COMEX - Crop mIoU: leaderboard

Metric: Mean crop IoU (0-1, native; overlap between the predicted crop box and the expert crop over the COMEX test set of expanded images; zero-shot). Source: arxiv.org. Saturation forecast: Around 2032. 15 models tracked.

Top models

#ModelScore
1Seed 1.80.51
2Seed-1.60.51
3MiniMax-M2.50.51
4Step3 VL 10B0.49
5Qwen 3.5 9B0.49
6Qwen 3 VL 8B0.49
7Qwen 3.5 4B0.48
8Qwen 3.5 2B0.48

Interactive version: theaggregate.ai/benchmark?slug=comex-crop-miou · How It Works · Data refreshed daily, snapshot 2026-09-26.