Dynamic Spatial Reasoning on SAT Synthetic
65.2AccuracyWorld2VLM-GRPO
Evaluation Results
| Method | Links | |
|---|---|---|
| World2VLM-GRPOSetting=HY-WorldPlay as WM2026.04 | 65.2 | |
| World2VLM-GRPOSetting=SVC as WM2026.04 | 59.2 | |
| World2VLM-SFTSetting=HY-WorldPlay as WM2026.04 | 57.2 | |
| Qwen2.5-VL-7B + WMSetting=MindJourney-style[55]2026.04 | 51.75 | |
| World2VLM-SFTSetting=SVC as WM2026.04 | 50 | |
| Qwen2.5-VL-7BSetting=Base VLM2026.04 | 39.6 |