Debris flows are a growing natural hazard due to climate change and population den- sity. To effectively assess this hazard, simulating field-scale debris flows at a reasonable computational cost is crucial. We enhance existing debris flows models by rigorously de- riving a series of depth-averaged shallow models with varying complexities describing the behavior of grain-fluid flows, considering granular mass dilatancy and pore fluid pressure feedback. The most complete model includes a mixture layer with an upper fluid layer and solves for solid and fluid velocity in the mixture and for the upper fluid velocity. Simpler models are obtained by assuming velocity equality in the mixture or single-layer descrip- tions with a virtual thickness. Simulations in a uniform configuration mimicking submarine landslides and debris flows reveal that these models are extremely sensitive to the rheology, the permeability (grain diameter), and initial volume fraction, parameters that are hard to measure in the field. Notably, velocity equality assumptions in the mixture hold true only for low permeability (corresponding to grain diameters d = 10−3 m). One-layer models results can strongly differ from those of the complete model, for example the mass can stop much earlier. One-layer models however provide a rough estimate of two-layer models when permeability is low, initial volume fraction is distant from critical, and upper fluid layer is very thin. Our work demonstrates the need of developing two-layer models accounting for dilatancy and accounting for an upper layer made either of fluid or grains.