> Why go to the expense of training your own model, especially when it will be inferior to state of the art models.

Uuh.. No? Especially of the training data, as in this case, is of better quality.