> "...it appears that not only is Mistral training a version of this so-called “Miqu” model that approaches GPT-4 level performance, but it may, in fact, match

> "...it appears that not only is Mistral training a version of this so-called “Miqu” model that approaches GPT-4 level performance, but it may, in fact, match or exceed it, if his comments are to be interpreted generously."