Multidisciplinary knowledge and reasoning on MMMU (dev)
25.33ScoreDefender Iter. 3
Evaluation results
| Method | Links | |
|---|---|---|
| Defender Iter. 3Training Strategy=Iterative Co-evolution2026.01 | 25.33 | |
| Defender Iter. 2Training Strategy=Iterative Co-evolution2026.01 | 23.33 | |
| Yang et al.Training Strategy=Finite Augmentation Datasets2026.01 | 22.67 | |
| Ablation (Clean Data)Training Strategy=Ablation2026.01 | 21.33 | |
| Base (M_def^(0))Training Strategy=Base2026.01 | 20.67 | |
| Liu et al. (Insert)Training Strategy=Finite Augmentation Datasets2026.01 | 20.67 | |
| Defender Iter. 1Training Strategy=Iterative Co-evolution2026.01 | 20 | |
| Liu et al. (Add)Training Strategy=Finite Augmentation Datasets2026.01 | 19.33 | |
| Liu et al. (All)Training Strategy=Finite Augmentation Datasets2026.01 | 18 |