PLUTO is a powerful framework designed to push the limits of imitation learning-based planning for autonomous driving. The framework addresses three key challenges: multi-modal driving behaviors, efficient auxiliary loss computation, and enhanced interaction learning through contrastive learning. PLUTO features a longitudinal-lateral aware model architecture, an innovative auxiliary loss computation method, and a novel training framework that leverages contrastive learning with new data augmentations. Evaluations using the nuPlan dataset demonstrate that PLUTO achieves state-of-the-art closed-loop performance, surpassing both learning-based and rule-based planners. The main contributions include a query-based model architecture for flexible driving behaviors, a novel auxiliary loss calculation method, and the Contrastive Imitation Learning (CIL) framework with new data augmentations. The results and code are available at <https://jchengai.github.io/pluto>.PLUTO is a powerful framework designed to push the limits of imitation learning-based planning for autonomous driving. The framework addresses three key challenges: multi-modal driving behaviors, efficient auxiliary loss computation, and enhanced interaction learning through contrastive learning. PLUTO features a longitudinal-lateral aware model architecture, an innovative auxiliary loss computation method, and a novel training framework that leverages contrastive learning with new data augmentations. Evaluations using the nuPlan dataset demonstrate that PLUTO achieves state-of-the-art closed-loop performance, surpassing both learning-based and rule-based planners. The main contributions include a query-based model architecture for flexible driving behaviors, a novel auxiliary loss calculation method, and the Contrastive Imitation Learning (CIL) framework with new data augmentations. The results and code are available at <https://jchengai.github.io/pluto>.