Your comment makes no sense. How and why would a local model be nerfed anyway...?

he's saying that he notices a difference between local (not nerfable) and hosted ones, so that it's not as likely to be just placebo