In-hand object reorientation has been extensively studied in robotic manipulation, yet it remains limited by slow execution and a tendency to become trapped in contact-constrained local optima when handling large or geometrically complex objects. These limitations often lead to prolonged manipulation time or failure on challenging shapes. We propose Toss2Reorient, a dynamic manipulation framework for in-air object reorientation through a toss-and-catch paradigm. By leveraging toss-and-catch motion together with coordinated hand–arm control, the method reduces the average execution time by 50.6% across 30 diverse objects in simulation. While dynamic manipulation introduces additional instability, wrist-mounted 6-axis force/torque (F/T) sensing helps stabilize interaction dynamics, with only a modest 1.8-point increase in macro-average drop rate relative to in-hand reorientation. The policy is trained using asymmetric Proximal Policy Optimization, where the critic has access to privileged object state information while the actor operates on partial observations. We evaluate the proposed framework on 30 diverse objects in simulation. Experimental results show substantially faster reorientation compared to conventional in-hand reorientation, particularly for large and geometrically complex objects. Furthermore, the learned policy demonstrates strong zero-shot generalization, successfully performing reorientation on unseen objects with different shapes and physical properties without additional training.