Abstract: In order to improve the model performance for the forecast of tropical cyclone (TC) intensity and track in the western North Pacific and South China Sea and provide TC forecasts in the northern Indian Ocean, the GRAPES_TYM has been upgraded to version 3.0 in August 2019. The vertical levels are increased from 50 to 68, and the model domain is enlarged to cover the northern Indian Ocean. The results show that the increase in vertical resolution can reduce the mean track error of 72—120 h TC forecasts and remarkably reduce the mean intensity error of 24—96 h forecasts by decreasing the negative bias in the forecast of severe typhoons and super typhoons. There is no significant difference in the mean track and intensity errors when the model domain is expended to the northern Indian Ocean, except that the simulated TC track to the north of 20°N is more sensitive to the model domain. The comparative analysis of the retrospective results from 2016 to 2018 and the predictions of NCEP-GFS and ECMWF shows that the average track error simulated by the GRAPES_TYM-v3.0 is close to that by NCEP-GFS, and the error is larger than that simulated by ECMWF. The intensity error is obviously smaller than that by NCEP-GFS and ECMWF. NCEP-GFS and ECMWF both have larger negative biases. In summary, the intensity error in the forecast of GRAPES_TYM is reduced remarkably through increasing the vertical levels from 50 to 68, and long term track forecast is more sensitive to the enlarged domain which covers the northern Indian Ocean.