Math Reasoning on DynaMath (WCA)
26.8Worst Case Accuracy (WCA)Qwen2.5-VL-7B-Instruct + Faithful-MR1
Evaluation results
| Method | Links | |
|---|---|---|
| Qwen2.5-VL-7B-Instruct + Faithful-MR1Backbone model=Qwen2.5-VL-7B-Instruct, Training Data=19.2K2026.05 | 26.8 | |
| Perception-R1-7BTraining Data=1.4K2026.05 | 25.2 | |
| Qwen2.5-VL-7B-Instruct + VPPOBackbone model=Qwen2.5-VL-7B-Instruct, Training Data=19.2K2026.05 | 25.2 | |
| Vision-R1-7BTraining Data=210K2026.05 | 24.6 | |
| Vision-SR1-7BTraining Data=56K2026.05 | 23.4 | |
| Qwen2.5-VL-7B-Instruct + GRPOBackbone model=Qwen2.5-VL-7B-Instruct, Training Data=19.2K2026.05 | 22.2 | |
| Qwen2.5-VL-7B-InstructBackbone model=Qwen2.5-VL-7B-Instruct2026.05 | 20.4 | |
| Qwen2.5-VL-3B-Instruct + Faithful-MR1Backbone model=Qwen2.5-VL-3B-Instruct, Training Data=19.2K2026.05 | 18.6 | |
| Qwen2.5-VL-3B-Instruct + VPPOBackbone model=Qwen2.5-VL-3B-Instruct, Training Data=19.2K2026.05 | 14.8 | |
| Qwen2.5-VL-3B-InstructBackbone model=Qwen2.5-VL-3B-Instruct2026.05 | 13.6 | |
| Qwen2.5-VL-3B-Instruct + GRPOBackbone model=Qwen2.5-VL-3B-Instruct, Training Data=19.2K2026.05 | 11.6 |