The coarse-grained discrete particle model (DPM) is fast growing into a powerful tool and a useful counterpart of the widely used two-fluid model (TFM) in simulation of large-scale reactors. This work aims to study the role of mesoscale modeling in both TFM and DPM approaches to understand the advantage and disadvantage of each approach for further development. Both simulation approaches with and without considering mesoscale structures in drag modeling are systematically investigated through simulations of an industrial diameter-transformed fluidized bed reactor with complex reactions. It is found that considering mesoscale drag can obviously improve the prediction in solid concentration for both approaches, and the effect of mesoscale drag for TFM modeling is more significant than for DPM approach. Besides, the DPM approach can reveal local heterogeneous structures without using mesoscale drag because it can distinguish different parcels in each fluid cell, but it overestimates the accumulation of solid particles below the distributor, as the large coarse-grain ratio may over-enhance the particle collision. For reaction, the coke content can be better predicted by both approaches with mesoscale drag, and the DPM simulation can capture more heterogeneous distribution of coke content than TFM modeling. The predicted temperature and product distribution still have obvious deviation from industrial data, suggesting a need of mesoscale heat and mass transfer modeling. The underlying mechanisms are further analyzed together with proposing future work.