Recent research indicated that processes as easy as two-digit number comparison are under cognitive control. In the present study we evaluated the validity of the cognitive control architecture proposed by Macizo and Herrera (2011, 2012) to account for their empirical findings regarding the unit-decade compatibility effect: longer responses for unit-decade incompatible (e.g., 37_52, 3<5, but 7>2) as compared to compatible number pairs (e.g., 42_57, 4<5 and 2<7). Artificial neural network models implementing a cognitive control circuit were trained for magnitude comparison on stimulus sets with different proportions of compatible and incompatible pairs (20%, 50% and 80% incompatible pairs). In line with the empirical observations of a proportion congruity effect, the simulated compatibility effect increased with the number of compatible pairs. Finally, with the same model specifications we were also able to simulate the Gratton effect, providing further evidence for cognitive control in number processing to be substantiated by future empirical studies.