In this study, large eddy simulations (LES) are conducted using both a finite-rate chemistry (FRC) model and a flamelet/progress-variable (FPV) model for a series of piloted partially premixed methane/air flames of increasing turbulence intensity (Sandia Flames D, E, and F). From Flame D to E to F, as flow velocity and strain rate increase, the flame is either pushed downstream and extended radially or weakened by enhanced local extinction. The two combustion models produce different spatial distributions of both time-averaged quantities and instantaneous flame field. The FPV model provides an overall better prediction of the time-averaged axial and radial profiles of Flame D, but a significantly worse prediction of Flame F, primarily because the FPV model significantly over predicts local extinction. In terms of the conditional statistics, in which the effects of spatial distribution of mixture fraction and subgrid-scale (SGS) modeling are largely "removed," the FRC model provides better predictions than the FPV model for all quantities at most locations and mixture fractions in all three flames. The effect of differential diffusion on the prediction of a species depends on the molecular diffusivity of that species; the effect is typically smaller than the difference between the FRC and FPV models.