Meta Motivo, released by Meta FAIR, is the first behavior-based model that leverages a novel unsupervised reinforcement learning algorithm for pre-training, designed to control complex virtual humanoid agents in completing full-body tasks. The model can tackle unseen tasks during testing, such as motion tracking, pose reaching, and reward optimization, without requiring additional learning or fine-tuning. The significance of this technology lies in its zero-shot learning ability, capable of managing a variety of complex tasks while maintaining behavioral robustness. The development of Meta Motivo stems from a pursuit of generalization capabilities for more complex tasks and different types of agents. Its open-source pre-trained models and training code encourage the community to further advance research on behavior-based models.