Xiaomi Auto has launched the Xiaomi EV World Model, a groundbreaking framework for autonomous driving that integrates 3D reconstruction with video generation. The JointWM architecture sets new performance records on benchmarks like Waymo and nuScenes by using 3D geometric structures to anchor scenes and predict unobserved regions. This approach allows for rapid video reconstruction and efficient generation, achieving speeds of 0.19 seconds per view on an H20 GPU. The framework's reconstruction module, WorldRec, uses sparse 3D query points to create a 4D Gaussian spatial skeleton, enabling quick scene reconstruction. The generation module, WorldGen, enhances visual details within physical boundaries and predicts future frames. This innovation has been deployed in Xiaomi Auto's synthetic data generation, simulation environments, and an Assistant Driving Academy, significantly advancing autonomous driving technology.