Spatiotemporal Reasoning on VSI-Bench
41.55AccuracyWorld2VLM-GRPO
Evaluation Results
| Method | Links | |
|---|---|---|
| World2VLM-GRPOSetting=SVC as WM2026.04 | 41.55 | |
| World2VLM-SFTSetting=SVC as WM2026.04 | 39.84 | |
| World2VLM-GRPOSetting=HY-WorldPlay as WM2026.04 | 39.07 | |
| World2VLM-SFTSetting=HY-WorldPlay as WM2026.04 | 38.63 | |
| Qwen2.5-VL-7B + WMSetting=MindJourney-style[55]2026.04 | 37.68 | |
| Qwen2.5-VL-7BSetting=Base VLM2026.04 | 33 |