The authors show that neural activity and synaptic plasticity in the orbitofrontal cortex mediate multiple timescales of reinforcement learning (RL) for meta-RL, which parallels a form of meta-RL in artificial intelligence.