SAS Enterprise Miner development is routinely implementing new features and options in our software. In some cases, rounding is handled at different levels of precision on different types of machines due to differences in hardware/OS. In other cases, default settings might have changed. If you are using non-linear models (tree-based models, neural network models, etc...) these are iterative methods which often involve randomization in their optimization. It is possible that slight differences might occur. In general, however, I would expect the predicted values to be similar unless there was a problem with how the earlier version of the software was handling a certain situation. Without a great deal more detail, there is no way to know for certain the cause of the differences you are inquiring about.