For the Japan Sodium-cooled Fast Reactor (JSFR), should the hypothesized core disruptive accident (CDA) happened, the in-vessel retention (IVR) will be the main target to achieve. In the heat-removal phase of the CDA, the debris bed will be piled up on the debris catcher. The capability of stable cooling and avoiding recriticality on the debris bed will be the main issues for achieving IVR. Previous studies have shown that the homogeneous debris bed can attain stable cooling and eliminate the probability of recriticality. Besides, self-leveling, which is a mechanism redistributing and flattening the debris bed by the natural circulation or vaporization from surrounding coolant, can further suppress the debris bed’s thickness to below the coolable thickness. However, in the real situation, the debris bed is composed of mixed-density debris particles. Hence, when these mixed-density debris particles start to redistribute due to self-leveling, the debris bed will form a heterogeneous density distribution. Under this scenario, the capability of coolability and the probability of recriticality could deviate from the previous study. Therefore, it is necessary to obtain a verified coupled model between the computational fluid dynamics (CFD) and the discrete element method (DEM) to track the mixed-density debris particles’ movement under the phenomenon of self-leveling. In this paper, first, the experiments simulating self-leveling on the mixed-density particle bed are performed. Afterward, the random heavy particle movement’s experimental data are extracted and transformed into the statistics form as the benchmark materials. Finally, the CFD-DEM model is validated via a series of sensitivity studies. The verified CFD-DEM can be expected to simulate the self-leveling behavior on the mixed-density debris bed and the real reactor case.