The cerebellar cortex contributes to diverse behaviors by transforming mossy fiber inputs into predictions in the form of Purkinje cell (PC) outputs, and then refining those predictions. Molecular layer interneurons (MLIs) account for approximately 80% of the inhibitory interneurons in the cerebellar cortex, and are vital to cerebellar processing. MLIs are thought to primarily inhibit PCs and suppress the plasticity of excitatory synapses onto PCs. MLIs also inhibit, and are electrically coupled to, other MLIs, but the functional significance of these connections is not known. Behavioral studies suggest that cerebellar-dependent learning is gated by disinhibition of PCs, but the source of such disinhibition has not been identified. Here we find that two recently recognized MLI subtypes, MLI1 and MLI2, have highly specialized connectivity that allows them to serve very different functional roles. MLI1s primarily inhibit PCs, are electrically coupled to each other, fire synchronously with other MLI1s on the millisecond time scale in vivo, and synchronously pause PC firing. MLI2s are not electrically coupled, they primarily inhibit MLI1s and disinhibit PCs, and are well suited to gating cerebellar-dependent learning. These findings require a major reevaluation of processing within the cerebellum in which disinhibition, a powerful circuit motif present in the cerebral cortex and elsewhere, greatly increases the computational power and flexibility of the cerebellum. They also suggest that millisecond time scale synchronous firing of electrically-coupled MLI1s helps regulate the output of the cerebellar cortex by synchronously pausing PC firing, which has been shown to evoke precisely-timed firing in PC targets.