The use CFD-DEM for the simulation of dense gas–solid flows comes with limitations based on the numerical cell size depending on the diameter of the particles. With the introduction of a multi-grid approach, i.e., the calculation of the fluid phase and the solid phase are done with two separate meshes, a finer and a coarser one, this limitation can be overcome. In this work, a three-level grid approach is proposed, in which an additional numerical grid is used to transform parameters used for the solid-fluid momentum exchange from the Lagrangian to the Eulerian grid. The resolution of this newly introduced grid lies in between the resolution of the two other meshes. A conventional single- and dual-grid, as well as the newly introduced three-level grid approach were validated with experimental data in literature and afterwards compared in terms of accuracy and simulation speed. A multi-grid approach, dual- or three-level, shows a higher accuracy than the single-grid method. An additional third grid achieves in almost all cases slightly better results than the dual-grid method while the increase in simulation time is negligible. A coarsening of the solid grid has a higher effect on accuracy than the refinement of the fluid grid. Furthermore, the coarse grain method was implemented to increase simulation speed. Still, the three-level grid approach requires a deeper understanding in the resolution of the numerical grid as there are more degrees of freedom for the cell sizes, but an adequate choice of the numerical grid can drastically improve the simulation accuracy and speed in combination with a coarse grain method.