Did you set the model code to gpt-4-0314?

I did, and I still get the original speed (produces tokens at about the speed you would read aloud), and I haven't seen quality change.