If Y1 and Y2 are just two different categories tiering of the exact same ultimate dependent variable Y; then I suspect some comparison of the two models may make some sense. If on the other hand Y1 and Y2 are two entirely different dependent variables (not only differentiated by the number of categories), then your question does not make sense. And, the straight answer is that those two models are simply not comparable.
Assuming that Y1 and Y2 are essentially derived from the exact same variable, you could compare the quality and the fit of the models using standard Goodness-of-fit measures for Probit model, and observe which categorie segmentation gave you better result. Or for the purpose of model parsimony, did increasing your number of categories from 6 to 7 was materially worth it from both an explanatory and statistical significance standpoint. This can be an interesting process, and you could test additional segmentation cuts of this same model structure using for instance fewer categories (like 4 or 5) and more categories (8, 9, etc.). Selecting the proper number of categories is based on both art and science (statistical significance of the difference between the various categories, etc.).