Moist adiabatic processes in the tropics amplify the surface warming, producing a warming peak at approximately 200 hPa, known as the “tropical tropospheric amplification”. Tropical tropospheric amplification, as a remarkable feature of climate change, is an important metric in evaluating model performances. In this study, based on RSS4.0 satellite data and ERA5.1 reanalysis data, we systematically assess the ability of the Flexible Global Ocean-Atmosphere-Land System Version 3 (FGOALS-g3) model in simulating temperature change, especially the tropical tropospheric amplification, and reveal improved simulation skills in the latest version, FGOALS-g3, compared with those in the previous version, FGOALS-g2. By comparing the results of the historical simulation of FGOALS-g3 with those of the simulation from its atmospheric component, the Grid-Point Atmospheric Model of LASG-IAP (GAMIL3), the role of air–sea coupling is studied. The results show that FGOALS-g3 can reasonably reproduce the observed significant global tropospheric warming, but with a stronger trend that is related to internal variability of the climate system and the differences in historical external forcing used by the two generations of climate system models. FGOALS-g3 has also appropriately simulated the observed vertical profile of mean tropical warming and spatial distribution of the tropical tropospheric amplification. This model has shown a positive bias in the simulated magnitude of the tropical tropospheric amplification, resulting from a greater temperature change in the lower troposphere. Compared with FGOALS-g2, the improvement in FGOALS-g3 is mainly manifested as an enhanced response to volcanic aerosol forcing, a more reasonable spatial pattern of the amplified tropical troposphere and the vertical profile of the mean temperature trend. The GAMIL3 simulation fails to influence the external forcing changes on the tropospheric warming trend because of a lack of the air–sea coupling, leading to biases in the long-term trend simulation. However, the GAMIL3 simulation reasonably captures interannual variability because it is driven by the observed sea-surface temperature.