DispatchGym is a research framework enhancing reinforcement learning (RL) applications for dispatch systems, specifically for Grab. It enables data scientists to develop and test RL solutions independently, addressing critical challenges like reward function sensitivity and lever influence. The framework's modular design in Python, complemented by Numba for performance, promotes rapid iteration and innovation in optimizing driver dispatching. DispatchGym’s emphasis on reliability and cost-effectiveness makes it accessible to users, driving impactful research in the RL domain for real-world applications.