Nah its because they cache and preprocess requests by dumb models and send them too often to another dumb models instead of the top tier model.