Compared to the Coupled Model Intercomparison Project Phase 5 (CMIP5) climate models, the Antarctic sea ice area (SIA) has been improved in Phase 6 (CMIP6). However, the lack of knowledge about the reliability of sea ice dynamic and thermodynamic processes in the CMIP6 models still limits the accuracy of Antarctic sea ice projections. Here, by using a novel and systematic statistical metric, the performance of CMIP5 and CMIP6 models with near-realistic SIAs was assessed. We found improvements in CMIP6 models relative to CMIP5. Moreover, forcing the sea ice-ocean model with atmospheric reanalysis led to excessive ice convergence compared to the fully coupled ocean-sea ice-atmosphere model, although the SIA bias could be much smaller. This prevalent insufficient ice divergence in the models is highly correlated with the negative ice thickness bias, highlighting the importance of ice thickness in the correct simulation of sea ice dynamics.