List your company

Video Reasoning on VBVR-Bench

97.4Overall AccuracyHuman
24.49643.42362.3581.277Mar 17, 2026Mar 25, 2026Apr 3, 2026Apr 12, 2026Apr 21, 2026Apr 30, 2026May 9, 2026

Evaluation results

MethodLinks
HumanType=Baseline2026.0397.4
VBVR-Wan2.2 + CollabVRVGM Cost (s)=10.912026.0575.7
VBVR-Wan2.2 + Training-Free EnsembleType=Video Reasoning Models, Ensemble=reasoning path ensembling2026.0371.6
VBVR-Wan2.2 + Pass@4VGM Cost (s)=14.802026.0570.7
VBVR-Wan2.2 + Pass@2VGM Cost (s)=7.402026.0569.4
VBVR-Wan2.2Type=Video Reasoning Models2026.0368.5
VBVR-Wan2.2VGM Cost (s)=3.702026.0567.1
VBVR-Wan2.2 + VideoTPOVGM Cost (s)=11.102026.0565
Sora 2Type=Proprietary Video Models2026.0354.6
Veo 3.1Type=Proprietary Video Models2026.0348
Runway Gen-4 TurboType=Proprietary Video Models2026.0340.3
Cosmos-Predict2.5 + CollabVRVGM Cost (s)=10.912026.0540.3
Wan2.2-I2V-A14BType=Open-source Video Models2026.0337.1
Kling 2.6Type=Proprietary Video Models2026.0336.9
LTX-2Type=Open-source Video Models2026.0331.3
Cosmos-Predict2.5VGM Cost (s)=3.702026.0530.8
CogVideoX1.5-5B-I2VType=Open-source Video Models2026.0327.3
HunyuanVideo-I2VType=Open-source Video Models2026.0327.3