Abstract Genetic variations affecting dopaminergic neuromodulation such as the DRD2/ANKK1 and the COMT Val158Met polymorphisms contribute to goal-directed behavior that requires a balance between stabilization and updating of current states and behaviors. Dopamine is also thought to be relevant for encoding of surprise signals to sensory input and adaptive learning. A link between goal-directed behavior and learning from surprise is therefore plausible. In the present fMRI study, we investigated whether DRD2 and COMT polymorphisms are related to behavioral responses and neural signals in the caudate nucleus and dlPFC during updating or stabilizing internal models of predictable digit sequences. To-be-detected switches between sequences and to-be-ignored digit omissions within a sequence varied by information-theoretic quantities of surprise and entropy. We found that A1 noncarriers and Val-carriers showed a lower response threshold along with increased caudate and dlPFC activation to surprising switches compared with A1-carriers and Met-homozygotes, whose dlPFC activity increased with decreasing switch surprise. In contrast, there were overall smaller differences in behavioral and neural modulation by drift surprise. Our results suggest that the impact of dopamine-relevant polymorphisms in the flexibility-stability trade-off may result in part from the role of dopamine in encoding the weight afforded to events requiring updating or stabilization.