The obvious point of a model that works like this is to see if you can get better prompt understanding. Increasing the resolution in a small model would decrease the capacity for prompt adherence.