UniTraj is a unified framework for scalable vehicle trajectory prediction that integrates multiple datasets, models, and evaluation criteria. The framework addresses challenges in data domain shifts and dataset size impacts on model performance. It unifies diverse datasets, including nuScenes, Argoverse 2, and WOMD, and supports various trajectory prediction models like AutoBot, MTR, and Wayformer. UniTraj provides a standardized data format, unified evaluation metrics, and flexible data processing to enable comprehensive research. The framework allows for cross-dataset and cross-city generalization studies, revealing that model performance drops when transferred to new domains. However, increasing dataset size and diversity significantly improves performance, leading to a new state-of-the-art result on the nuScenes dataset. The framework also offers insights into dataset characteristics, showing that both size and diversity contribute to model generalization. UniTraj enables extensive experiments, demonstrating that larger, more diverse datasets enhance model performance and generalization. The framework is released as an open-source tool to foster further advancements in vehicle trajectory prediction. Key contributions include the introduction of UniTraj, investigation of model generalization across datasets and cities, exploration of data scaling impact, and in-depth dataset analysis. The framework provides a comprehensive platform for trajectory prediction research, supporting diverse research questions and experimental setups.UniTraj is a unified framework for scalable vehicle trajectory prediction that integrates multiple datasets, models, and evaluation criteria. The framework addresses challenges in data domain shifts and dataset size impacts on model performance. It unifies diverse datasets, including nuScenes, Argoverse 2, and WOMD, and supports various trajectory prediction models like AutoBot, MTR, and Wayformer. UniTraj provides a standardized data format, unified evaluation metrics, and flexible data processing to enable comprehensive research. The framework allows for cross-dataset and cross-city generalization studies, revealing that model performance drops when transferred to new domains. However, increasing dataset size and diversity significantly improves performance, leading to a new state-of-the-art result on the nuScenes dataset. The framework also offers insights into dataset characteristics, showing that both size and diversity contribute to model generalization. UniTraj enables extensive experiments, demonstrating that larger, more diverse datasets enhance model performance and generalization. The framework is released as an open-source tool to foster further advancements in vehicle trajectory prediction. Key contributions include the introduction of UniTraj, investigation of model generalization across datasets and cities, exploration of data scaling impact, and in-depth dataset analysis. The framework provides a comprehensive platform for trajectory prediction research, supporting diverse research questions and experimental setups.