A regional-scale groundwater flow model covering 25,000 km² was built on a high-resolution (120 m) three-layer grid representing Quaternary sediments, with the domain divided according to mapped Newmarket-era sediment thicknesses that cover only portions of the model domain. Four MODFLOW 6 configurations were evaluated: a structured (baseline) grid, a structured grid that incorporates pass-through cells, the baseline grid only built using an unstructured mesh, and a “coarsened” unstructured grid in which the Newmarket unit is represented on a coarser mesh. The steady-state baseline model was calibrated against more than 148,000 hydraulic-head observations. The structured baseline, the structured grid with pass-through cells, and the fine unstructured grid all produced virtually identical steady-state heads, well-field drawdowns, and groundwater-surface-water exchanges, with differences far smaller than the calibration error. By contrast, the coarsened unstructured grid achieved almost a three-fold reduction in run time but showed the greatest deviation in steady-state head and seasonal flux. These results illustrate the trade-off between the simplicity of implementing structured models and the computational gains offered by unstructured discretization. Future work should explore variable-coarsening strategies, coarse-grid calibration techniques, and localized error analyses to support the efficient design of large regional groundwater models.