Multi-discipline Multimodal Reasoning on MMMU Pro
36.63Avg@8 AccuracyPAPO_G
Evaluation Results
| Method | Links | |
|---|---|---|
| PAPO_GBackbone=Qwen2.5-VL, Model Size=7B2025.07 | 36.63 | |
| PAPO_DBackbone=Qwen2.5-VL, Model Size=7B2025.07 | 36.34 | |
| GRPOBackbone=Qwen2.5-VL, Model Size=7B2025.07 | 35.17 | |
| DAPOBackbone=Qwen2.5-VL, Model Size=7B2025.07 | 29.02 | |
| PAPO_DBackbone=Qwen2.5-VL, Model Size=3B2025.07 | 28.76 | |
| DAPOBackbone=Qwen2.5-VL, Model Size=3B2025.07 | 28.42 | |
| PAPO_GBackbone=Qwen2.5-VL, Model Size=3B2025.07 | 27.11 | |
| GRPOBackbone=Qwen2.5-VL, Model Size=3B2025.07 | 25.66 | |
| PAPO_GBackbone=Qwen3-VL (thinking), Model Size=2B2025.07 | 23.42 | |
| GRPOBackbone=Qwen3-VL (thinking), Model Size=2B2025.07 | 20.51 |