Rapid urbanization in the past decades has significantly escalated mobility de-mands and imposed higher service quality standards. As a promising solution to address this challenge, Autonomous Mobility-on-Demand (AMoD) systems offer tailored mobility services while facilitating centralized control. In this paper, we propose a reinforcement learning-based AMoD control strategy based on joint ve-hicle rebalancing and dynamic trip pricing. By leveraging the inductive biases of graph neural networks and a bi-level framework that incorporates optimization rou-tines in the inner loop to reduce the action dimension, we are able to significantly improve model scalability, sample efficiency, and convergence speed. Through ex-periments conducted with real-world data from New York City and San Francisco, we demonstrate the effectiveness of the proposed joint pricing and rebalancing pol-icy compared to dynamic pricing or rebalancing policies alone. An analysis of the behavior of the policy reveals that the two control mechanisms can work in tandem to effectively address the limitations inherent to independent dynamic pricing or rebalancing policies under different scenarios. Crucially, the learned policies are shown to generalize well to other unseen scenarios and cities. Furthermore, we extend our framework to offline reinforcement learning, thus allowing us to learn policies from static historical datasets that can achieve performance comparable to online learning. The success of offline reinforcement learning underscores the potential of our proposed framework as a viable solution for controlling real-world risk-conscious transportation systems.