AbstractEquilibrium climate sensitivity (ECS) and its related feedbacks are important metrics that can be used to measure the global mean surface temperature change in future climate projections, and the cloud feedback (CF) is considered to be the main contributor to ECS uncertainties. However, the use of different CF methods significantly affects the CF amplitudes, and the sign of the CF may also change. Combining the radiative kernel approach with a simplified CF method, the differences in the ECS and associated feedbacks between two versions of the Flexible Global Ocean–Atmosphere–Land System model (i.e., FGOALS‐g2 and FGOALS‐g3) were analyzed. Results show that the ECS of FGOALS‐g3 is smaller than that of FGOALS‐g2 (2.8 vs. 3.3 K). The main feedback processes that contribute to the ECS change in FGOALS‐g3 are the weaker surface albedo feedback and stronger negative shortwave CF. The reduced surface albedo feedback in FGOALS‐g3 can be attributed mainly to the differences in the simulations of sea ice area, surface temperature, and the Atlantic Meridional Overturning Circulation, as well as their interactions, compared with FGOALS‐g2. The enhanced negative shortwave CF in FGOALS‐g3 is directly related to the strengthened feedback of cloud area fraction and liquid water path. Furthermore, these changes can be traced back to the different atmospheric moist processes, parameter tuning, ocean grid, and external forcings used in FGOALS‐g3, as these all affect the mean climate state of the model.