An Order-Aware Mixed-Palletizing (OAMP) framework for New Retail is proposed to address problems of frequent offline handling and limited observation space in the full-pallet outbound mode.Mixed-Palletizing for New Retail was decoupled into two sub-tasks of outbound decision-making and palletizing decision-making through the collaboration between the decision layer and the physical layer. An Outbound Decision-Making (ODM) algorithm was designed for the outbound decision-making stage. The outbound sequence of items in the order was dynamically adjusted through an energy function that quantified the geometric matching between candidate goods and the remaining pallet space. An online 3D bin-packing algorithm based on deep reinforcement learning was integrated into the palletizing decision-making stage to drive the robotic arm for precise placement. Simulation experiments were conducted in different scenarios. It is shown that average space utilization is increased by approximately 5.38% compared to traditional heuristic rules while a real-time response of approximately 50 ms per box is maintained. High-efficiency technical support is provided for the upgrade of new retail warehousing systems.