This paper presents a scalable and practical generative model for autonomous driving, named GUMP (Generative Unified Model for Motion Planning). GUMP is designed to learn the dynamics of driving scenes, enabling it to simulate diverse future scenarios and generate various driving scenarios based on prompts. The model operates in both full-Autoregressive and partial-Autoregressive modes, improving inference and training speed without sacrificing generative capability. GUMP is evaluated on two real-world datasets: the Waymo motion dataset and the nuPlan dataset, achieving state-of-the-art performance in simulation realism and scene generation benchmarks. In planning benchmarks, GUMP outperforms prior methods. The model can be used as a data generator, realistic simulator, planner, and online training module, making it a versatile foundation for motion planning tasks in autonomous driving. The source code is publicly available at <https://github.com/HorizonRobotics/GUMP/>.This paper presents a scalable and practical generative model for autonomous driving, named GUMP (Generative Unified Model for Motion Planning). GUMP is designed to learn the dynamics of driving scenes, enabling it to simulate diverse future scenarios and generate various driving scenarios based on prompts. The model operates in both full-Autoregressive and partial-Autoregressive modes, improving inference and training speed without sacrificing generative capability. GUMP is evaluated on two real-world datasets: the Waymo motion dataset and the nuPlan dataset, achieving state-of-the-art performance in simulation realism and scene generation benchmarks. In planning benchmarks, GUMP outperforms prior methods. The model can be used as a data generator, realistic simulator, planner, and online training module, making it a versatile foundation for motion planning tasks in autonomous driving. The source code is publicly available at <https://github.com/HorizonRobotics/GUMP/>.