No, it looks like the stance of the FSF is that models should be free as a matter of principle, the same as their stance when it comes to software. Nothing in the linked post contradicts the description that the judgement was that the training was fair use.