The Arctic wintertime energy and moisture budgets are largely controlled by the advection of warm, moist air masses from lower latitudes; the cooling and drying of these air masses inside the Arctic; and the export of cold, dry air masses. Climate models have substantial difficulties in representing key processes in these air- mass transformations, including turbulence under stable stratification and mixed-phase cloud processes. Here, we use radiosonde profiles of temperature and moisture and surface radiation observations from Ny-Ålesund, Svalbard (1993–2014), to assess the properties of air masses being imported into and exported from the central Arctic in CMIP6 climate models. In the free troposphere, models tend to be cold-biased, especially for the coldest temperatures, and relative humidity in most models is closer to saturation with respect to ice than what is observed. In the analysed models, supersaturation with respect to ice tends to be better represented with two-moment microphysics. The overall distribution of column-integrated precipitable water in models matches well with observations. Cold and dry biases are stronger in air masses being exported from the Arctic than those entering the Arctic. This suggests that the previously reported cold bias in the Arctic in CMIP6 models is probably due to errors in local thermodynamic processes.