Hello Rachel,
First, developing the "best" regression model, whether OLS or GLM, is more of an art than a science. Therefore, one should not rely upon a single best fit statistic. I generally examine several statistics and graphs, etc. For example, I compare the r-square and adjusted r-square and standard error and checks for whether the model assumptions have been met, and whether the coefficient values and predicted values make sense (i.e., have face validity).
Also, there are many types of GLM models. So one would need to consider which type of GLM models you have developed in order to provide you with the best advice. I recently developed a generalized linear multiple regression model, and to evaluate the best fit for my model I used the same statistics and graphs that I would use for evaluating an OLS model. However, had my model instead involved logistic regression I would have varied my reference checks from those just mentioned.
Good luck.
------------------------------
Gretchen Donahue
------------------------------