Both model-based and model-free reinforcement learning methods are useful and sometimes are combined in an agent.