| Paper | Code | VBench-2.0 Total Score | VBench-2.0 Physics Score | VBench-2.0 Creativity Score | VBench-2.0 Commonsense Score | VBench-2.0 Human Fidelity Score | VBench-2.0 Controllability Score | ModelName | ReleaseDate |
|---|---|---|---|---|---|---|---|---|---|
| LightVAE: Towards Compact and Efficient Video Autoencoders | ✓ Link | 66.5546 | 75.3712 | 45.4439 | 74.0825 | 82.6056 | 55.2702 | LightVAE-Lite (MiniMax-H3) | 2026-10-10 |
| LongCat-Video Technical Report | ✓ Link | 62.11 | 59.92 | 54.73 | 70.94 | 80.20 | 44.79 | LongCat-Video | 2025-10-25 |
| TC-Pade: Trajectory-Consistent Pade Approximation for Diffusion Acceleration | 60.38 | TC-Padé (fast) | 2026-03-03 | ||||||
| PyramidalWan: On Making Pretrained Video Model Pyramidal for Efficient Inference | 57.48 | Wan-DMD (4 steps) | 2026-01-08 | ||||||
| ReHyAt: Recurrent Hybrid Attention for Video Diffusion Transformers | 56.3 | ReHyAt (15×Tc=5) | 2026-01-07 | ||||||
| LatSearch: Latent Reward-Guided Search for Faster Inference-Time Scaling in Video Diffusion | ✓ Link | 55.25 | 53.41 | 57.53 | 57.36 | 84.01 | 23.96 | LatSearch | 2026-03-15 |
| Breaking the Uniformity Trap: Scaling Video Diffusion Model via SplitMoE | 69.41 | 58.46 | 64.89 | 84.47 | 45.72 | SplitMoE | 2026-09-29 |