for you, I am in general not constrained by the speed of models since I parallelize. For me autonomy and accuracy are paramount above all else.