Abstract: Using the analyses and leading 35-day predictions with the Global and Regional Assimilation Prediction System-Global Forecast System (GRAPES-GFS) during the period from September 2018 to August 2019, we diagnosed the prediction errors and evaluated the extended forecast capability to provide a numerical weather guidance for the prediction at a sub-seasonal timescale. Results showed that, GRAPES-GFS could capture the spatial distribution characteristics of 2-m temperatures and 500 hPa geopotential heights during the winter in 2018 and summer in 2019, however there existed large system bias related to 2-m temperature analysis in the desert plateau areas where there was significant thermal forcing effect, especially in arid areas of Africa. Related to the 2-m temperature, the Root-Mean-Square Errors (RMSE) of the leading 1- to 3-week predictions approximated to the linear growth. GRAPES-GFS possessed a high prediction skill in the East Asia and Austria but had relatively low prediction skills in the ocean areas compared with that of the land areas. For the leading 1- to 3-week predictions related to the 500 hPa geopotential height, the prediction skills were higher at the low latitudes than at the high latitudes of East Asia. Also, the prediction skills for the tropics were much lower than for the other regions, of which the northern hemisphere was higher than that of the southern hemisphere. Regarding to the related Madden-Julian Oscillation (MJO), it is found that GRAPES-GFS could reproduce the propagation characteristics of spatial-temporal variations related to the upper and lower zonal wind and could capture the location of strong convective activity signals. However, the positive anomaly of the Outgoing Long Wave Radiation (OLR) was much weaker and the negative anomaly was much stronger. GRAPES-GFS could skillfully forecast MJO with 11 leading days from the view of Anomaly Correlation Coefficient (ACC), which was about the same level as the results from other forecasting models. For the selected two strong MJO cases, GRAPES-GFS could describe the MJO propagation process exactly but had a stronger signal during the MJO developing and decaying periods.