Comment by ElFitz

There will always be dataset and training quirks, and the provider’s own biases and focus, granting one model an edge over the others in some specific domain.