I don’t think it’s available for custom GPTs yet, but I hope that becomes an option soon, because when they made it the default model for a couple of days, it seemed to work really well with my GPTs. Though I’ve managed to replicate it somewhat on the default chat with some fiddling around.
But if they’re smart, they’ll allow users to choose which model they want for their GPTs, since different ones will work better with different models.