Comparing models' responses