Simulating granular flows across solid-like and fluid-like regimes, especially in complex scenarios like landslides, involves large deformations and strong nonlinearity. Accurately predicting the full dynamics requires careful implementation, verification and benchmarking at both constitutive and governing equation levels. The Material Point Method (MPM) has emerged as a robust approach to modeling granular flow problems. HydraxMPM is a high-performance, JAX-based MPM solver, which addresses these challenges. It utilizes JAX, a machine learning library optimized for high-performance array-based computation, to provide features such as automatic differentiation (AD). These capabilities enable efficient constitutive modeling and support gradient-based optimization. In this study, HydraxMPM is applied to various benchmark cases and its results are compared against reference data at the levels of constitutive laws and governing equations. Benchmark cases include material behavior under constant pressure shear and granular flow dynamics during column collapse. This work highlights the versatility of JAX in facilitating modularity of the code structure, diagnosis tasks such as sensitivity analysis, and simulations for dense granular flows.