I've been using 4.6 models since each of them launched. Same for 4.5.
4.6 performers worse or the same in most of the tasks I have. If there is a parameter that made me use 4.6 more frequently is because 4.5 get dumber and not because 4.6 seemed smarter.