The ability to cope with changing environments is critical for healthy functioning, yet this flexibility is impaired in many neuropsychiatric disorders. However, neural mechanisms underlying flexible behavior remain elusive. Here, we report that oscillatory dynamics in the medial prefrontal cortex (mPFC) support learning to flexibly overcome established behavioral bias. Mice performed a delayed non-match-to-sample task that required trial-by-trial adjustment of arm choice strategy despite persistent arm bias. Decoding analysis of delay-period local field potentials (LFPs) and single-unit activities revealed evolving neural representations across trials as mice adapted to the task. Notably, mPFC neurons modulated by theta (4~12 Hz) bursts selectively encoded upcoming choice information after acquiring the new rule. In contrast, beta (12~30 Hz) bursts correlated with perseverative behavior and appeared to inhibit theta-modulated neuronal firing in mice showing adaptive behavior. These theta and beta bursts were temporally separated over the delay period, reflecting a dynamic gating mechanism. Thus, beta bursts shape neuronal ensembles that are modulated by theta bursts to facilitate flexible learning. This dynamic interaction provides a mechanistic basis for cognitive flexibility and provides insights into cognitive rigidity seen in neuropsychiatric disorders such as schizophrenia and autism.