This study compares various assumptions in different models to assess their capabilities to model vehicle-track interactions up to 2 kHz at a single rail-top defect. Field measurement data are used to evaluate discrepancies. The characteristics of contact force and axle-box acceleration (ABA) are first identified and qualitatively correlated with track, wheelset and contact models. Subsequently, the results from different models are quantitatively compared in terms of their capabilities to reproduce those characteristics. It is found that the differences in sleeper and wheel-rail contact models lead to the most significant discrepancies. The causes and physical implications of the quantified discrepancies are also discussed.