Multimodal Large Language Model Evaluation on MLLM Evaluation Suite (test)
74.94MMEReVision-2M
Evaluation Results
| Method | Links | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| ReVision-2MData=2M unpaired text samples2026.02 | 74.94 | 36.4 | 76.35 | 45.23 | 33.49 | 29.59 | 26.8 | 24.38 | 80.14 | 72.18 | 48.26 | 49.75 | 0.74 | |
| w/. ImageData=1M ground-truth image-text pairs, Role=Upper bound baseline2026.02 | 73.59 | 35.4 | 76.01 | 44.18 | 34.8 | 27.7 | 25.9 | 23.27 | 80.87 | 69.13 | 47.11 | 48.91 | 1 | |
| ReVision*Data=1M unpaired text samples2026.02 | 72.2 | 34.33 | 75.84 | 43.72 | 30.22 | 27.64 | 25.7 | 21.03 | 79.59 | 71.93 | 46.37 | 48.05 | 0.37 | |
| UnicornData=1M unpaired text samples, Alignment Strategy=Mean shift2026.02 | 60.24 | 35.13 | 68.81 | 42.35 | 36.87 | 34.05 | 26.8 | 29.53 | 42.32 | 64.21 | 43.01 | 43.94 | 3.98 |