We investigate the ability of modern general circulation models (GCMs) to simulate transport in the martian atmosphere using measurements of argon as a proxy for the transport processes. Argon provides the simplest measure of transport as it is a noble gas with no sinks or sources on seasonal timescales. Variations in argon result solely from ‘freeze distillation’, as the atmosphere condenses at the winter poles, and from atmospheric transport. Comparison of all previously published models when rescaled to a common definition of the argon enhancement factor (EF) suggest that models generally do a poor job in predicting the peak enhancement in southern winter over the winter pole – the time when the capability of the model transport approaches are most severely tested. Despite observed peak EF values of ∼6, previously published model predictions peaked at EF values of only 2–3. We introduce a new GCM that provides a better treatment of mass conservation within the dynamical core, includes more sophisticated tracer transport approaches, and utilizes a cube–sphere grid structure thus avoiding the grid-point convergence problem at the pole that exists for most current Mars GCMs. We describe this model – the Ashima Research/Massachusetts Institute of Technology Mars General Circulation Model (Ashima/MIT Mars GCM) and use it to demonstrate the significant sensitivity of peak EF to the choices of transport approach for both tracers and heat. We obtain a peak EF of 4.75 which, while over 50% higher than any prior model, remains well short of the observed value. We show that the polar EF value in winter is primarily determined by the competition between two processes: (1) mean meridional import of lower-latitude air not enriched in argon and (2) the leakage of enriched argon out of the polar column by eddies in the lowest atmospheric levels. We suggest possibilities for improving GCM representation of the CO2 cycle and the general circulation that may further improve the simulation of the argon cycle. We conclude that current GCMs may be insufficient for detailed simulation of transport-sensitive problems like the water cycle and potentially also the dust cycle.