A long-standing and pervasive problem in climate modeling is the proper representation of cloud albedo over the Southern Ocean (SO). In this study, we investigate the causes of SO cloud albedo biases using phase 6 of the Coupled Model Intercomparison Project (CMIP6) simulations and a cloud-controlling factor approach on daily time scales. Cloud albedo, computed from upwelling and downwelling shortwave radiation at the surface and top of the atmosphere, is averaged into bins of vertical velocity, surface wind, and sea surface temperature. The performance of 15 models in both atmosphere-only and ocean-coupled configurations is evaluated against Clouds and the Earth’s Radiant Energy System (CERES) satellite retrievals in combination with ERA5 reanalysis for the 2000–14 period. We find that the SO cloud biases maximize in the 508–658 oceanic band and that models tend to underestimate SO cloud reflectivity for cold conditions and weak surface winds. In turn, descent (ascent) conditions are consistently underestimated (overestimated) across models for both hemispheres in the same latitude band. With a very similar approach, we evaluate how models represent the observed cloud albedo hemispheric asymmetry over oceans, which also maximizes in the 508–658 band. The sign of the asymmetry is consistently predicted by all models, many of which also predict a similar magnitude to observations. However, this is also a consequence of compensating global biases as individually most models tend to either overpredict or underpredict cloud albedo in both hemispheres. We propose that surface temperatures less than 48C are important in explaining SO bias in cloud albedo and that they also partly explain the observed hemispheric asymmetry. Further, we hypothesize that the 48C threshold sets a hemispheric asymmetry in cloud phase.