Image Reasoning on MathVista
76.8AccuracyCoPD
Evaluation Results
| Method | Links | |
|---|---|---|
| CoPDTraining Strategy=CoPD2026.04 | 76.8 | |
| Mixed RLVRTraining Strategy=Mixed RLVR2026.04 | 75.55 | |
| Image-ExpertTraining Strategy=Image-Expert2026.04 | 75.1 | |
| Video-ExpertTraining Strategy=Video-Expert2026.04 | 74.85 | |
| MOPDTraining Strategy=MOPD2026.04 | 74.75 | |
| Text-ExpertTraining Strategy=Text-Expert2026.04 | 73.9 | |
| BaseTraining Strategy=Base2026.04 | 73.2 | |
| Scaled base VLM 7B + Joint FFTTrain Data=3.6M + 1.6M2026.06 | 49.7 | |
| Scaled base VLM 7B + MERIT (Proposed, 2D)Train Data=3.6M + 1.6M2026.06 | 43 | |
| Scaled base VLM 7BTrain Data=3.6M2026.06 | 39.4 | |
| Base VLM 7B + MERIT (Proposed, 2D)Train Data=0.7M + 1.6M2026.06 | 35.4 | |
| LLaVA-1.5-13B + Joint FFTTrain Data=0.7M + 0.2M2026.06 | 34.4 | |
| Base VLM 7B + Joint FFTTrain Data=0.7M + 1.6M2026.06 | 32.9 | |
| LLaVA-1.5-13BTrain Data=0.7M2026.06 | 27.7 | |
| Base VLM 7BTrain Data=0.7M2026.06 | 26.9 | |
| LLaVA-1.5-7BTrain Data=0.7M2026.06 | 25.6 | |
| LLaVA-7BTrain Data=0.6M2026.06 | 25.4 |