Meta-World provides a collection of continuous control robotic manipulation tasks designed to benchmark multi-task and meta-learning reinforcement learning algorithms. Built on Mujoco for physics simulation and following the Gymnasium API standard, it offers standardized benchmarks (MT1, MT10, MT50, ML1, ML10, ML45) to evaluate different aspects of RL algorithms including task generalization, few-shot learning, and transfer capabilities. The repository is maintained by the Farama-Foundation.