List your company

Multimodal Spatial Reasoning on Winoground

70.47AccuracySpecFlow
64.240465.857767.47569.0923Jun 1, 2026

Evaluation results

MethodLinks
SpecFlowFLOPs(G)=18390.14, Lat.(s)=0.49, Mem.(GB)=46.942026.0670.47
VoCoTFLOPs(G)=28092.92, Lat.(s)=0.80, Mem.(GB)=62.232026.0670.09
CODIFLOPs(G)=17985.55, Lat.(s)=0.50, Mem.(GB)=50.212026.0668.91
PCCoTFLOPs(G)=17556.18, Lat.(s)=0.49, Mem.(GB)=53.332026.0667.49
SparseVLMFLOPs(G)=23738.55, Lat.(s)=0.79, Mem.(GB)=52.882026.0666.01
LightFastVFLOPs(G)=25253.62, Lat.(s)=0.69, Mem.(GB)=56.752026.0665.83
HeimaFLOPs(G)=18259.34, Lat.(s)=0.50, Mem.(GB)=43.092026.0665.64
SoTFLOPs(G)=26639.10, Lat.(s)=0.74, Mem.(GB)=60.162026.0664.48