At a dedicated session, Baidu Apollo unveiled Apollo World, an agent-driven omni-modal world model for physical AI, marking its first public introduction. The company also proposed the "2+3" criteria for world models. The model has been deployed in global testing and validation for Apollo Go, helping scale up autonomous driving. Apollo World simulates all traffic participants in virtual scenarios as AI agents endowed with decision-making and interaction capabilities, enabling autonomous vehicles to travel between any origin and destination within a virtual city. Its technology stack can be extended to broader physical AI applications such as robots, achieving exceptionally high efficiency in virtual road testing. By leveraging real-world operational data, it establishes a closed loop in which real operations and virtual validation reinforce each other—helping absorb testing costs and deployment risks, break the "impossible triangle" of scaling autonomous driving, and drive industry development.