Objectives  In order to test and analyze the accuracy of the latest global pressure and temperature model (GPT3 model), we choose 18 IGS stations in China for test.

  Methods  The meteorological data in 2015-2017 from Global Geodetic Observing System (GGOS) Atmosphere and the tropospheric delay data in 2015 provided by International GNSS Service (IGS) are used to test the model meteorological parameters and the zenith tropospheric delay (ZTD). And the error characteristics of GPT3 model are analyzed jointly with other GNSS stations all over the world.

  Results  The results show that the accuracy and the stability of GPT3 model are significantly improved compared with GPT and GPT2 models. The accuracy of GPT3 model is as same as that of GPT2w model and the stability is slightly improved. The GPT3 model is obviously sensitive to latitude change. In addition to water vapor pressure, the accuracy and stability of the model for estimating pressure and temperature decrease with the increase of latitude. The stability of water vapor pressure in the mid-latitude area and part of the low-latitude area is worse than that in the high-latitude area. The bias of GPT3 model is random in the low-altitude area, and the pressure bias is more obvious. When the altitude increases, the bias of pressure and water vapor pressure gradually stabilizes within ±2 hPa, and the temperature bias within ±2 ℃.

  Conclusions  GPT3 model shows strong stability and accurate prediction results than previous GPT models.