The classical verification of numerical weather prediction is assessment of forecast models accuracy over the whole spectrum of meteorological conditions observed in the atmosphere. Nevertheless, for operational forecasting it is important to understand the limitations of forecast data applicability in different situations. This issue can be solved within conditional verification by stratifying the dataset using synoptic typing. The research is devoted to the quality assessment for 2-meter air temperature forecasts based on the GFS global numerical weather prediction (NWP) model in the Perm region. Additionally, the forecast quality of the GFS and WRF-ARW models is compared for the summer season. The quality of forecasting is assessed across the entire sample of cases and sets of various synoptic situations. The best quality of forecasts is observed in autumn. Systematic underestimation of the predicted temperature is noticed in central parts of anticyclones in winter and in warm sectors of cyclones in spring