3 ms·

Actually it depends on the task. For many tasks, a smaller model can handle it, and it gets there faster!