Abstract: The Global and Regional Assimilation and Prediction System(GRAPES) model is set up as a new generation multi-scale universal data assimilation and numerical prediction system in China. The global forecasting system version of GRAPES_GFS 2.0 is formally established in June 2016, and thus a comprehensive assessment on its forecasting capacity is urgently needed. Comparing with NCEP FNL data, the hindcast of a whole year of 2014 and 4 seasonal representative months by GRAPES_GFS 2.0 are analyzed.The systematic error of 500 hPa potential height field is characterized by the obvious gradient and zonal distribution or wave columnar distribution, concentrated in mid and high latitudes. The error shows significant seasonal variation, which is much larger in winter than that in summer of both the north and south hemispheres. Furthermore, compared with the linear growth rate, GRAPES_GFS 2.0 forecast error is lower, and changing trends of errors with height are similar when lead time changes. The distribution of the initial error of 500 hPa temperature field is concentrated in tropics, while along with the growth of the forecast time, the large area of forecast error gradually moves to middle and high latitude areas. Moreover, the zonal average temperature error is mainly negative, while slightly positive near the tropopause of high latitude areas. There is no obvious distribution law of latitudinal wind field error, which is not closely related to latitude, sea land distribution and topography, alternated with west wind error and east wind error. The error of the height field in the tropopause, the temperature field and zonal wind field in the boundary layer and in the tropopause increases rapidly as well.Results above show that the evaluation on the oblique pressure instability of geopotential height field in mid and high latitudes still needs improving. As the low latitude area is dominated by positive pressure structure, the absolute error value with its growth is relatively small. Over-estimated thermal forces in plateau and desert regions result in the large error area of temperature field. The zonal wind field error is similar but may result in meridional wind error. In addition, the performance of the model in boundary layer and tropopause simulation needs improving.