vapi has an enterprise account with them so you get some cost savings, they also colocate / locally host models so inference can be faster