ManifoldActiveMULTIPLE_CHOICE · cpmm-multi-1

What is the main reason behind GPT-4o speed improvement relative to GPT-4 base model?