To gain insights into neural mechanisms enabling behavioral adaptations to complex and multidimensional environmental dynamics, we examined roles of VIP neurons in mouse medial prefrontal cortex (mPFC) in probabilistic reversal learning. Behaviorally, manipulating VIP neuronal activity left probabilistic classical conditioning unaffected but severely impaired reversal learning. Physiologically, conditioned cue-associated VIP neuronal responses changed abruptly after encountering an unexpected reward. They also conveyed strong reward prediction error signals during behavioral reversal, but not before or after, unlike pyramidal neurons which consistently conveyed error signals throughout all phases. Furthermore, the signal’s persistence across trials correlated with reversal learning duration. These results suggest that mPFC VIP neurons play crucial roles in rapid reversal learning, but not in incremental cue-outcome association learning, by monitoring significant deviations from ongoing environmental contingency and imposing error-correction signals during behavioral adjustments. These findings shed light on the intricate cortical circuit dynamics underpinning behavioral flexibility in complex, multifaceted environments.