CrossDim Intelligence has introduced UNITAS, the world's inaugural robot task-native 3D world model. This innovative model seamlessly integrates robot observations, actions, and environmental changes within a three-dimensional physical framework. By simultaneously modeling action and scene dynamics through 3D point trajectories, UNITAS effectively tackles the challenge of aligning heterogeneous data from diverse sources and trajectory parameters. Despite its compact size of only 1.7 billion parameters, UNITAS outperforms larger-scale comparable models across various robotic manipulation benchmarks, attaining an impressive average success rate of 85% in real-world robotic tasks. Ablation experiments further underscore the performance benefits derived from its 3D-native architecture. The accompanying lightweight module, TaskAnchor, enriches the robot's historical context awareness, bolstering its ability to execute long-term tasks. Furthermore, this technical methodology can be synergized with a self-developed simulation engine to construct a Physical AI data flywheel, offering a novel approach for scalable development of embodied AI grounded in unified physical world representations.