We review the coupled ice-ocean modeling activities aimed at predictions, both in the near term (days to a week) and in the long term (seasonal to decadal) of the polar oceans. First the state of the knowledge of potential predictability is exposed, then an overview is given of the tools available for carrying out such predictions: the observations that can be used to initialize actual predictions, the coupled ice-ocean–modeling, including the fully-coupled Earth System Models for long-term predictions, and data-assimilation techniques. Finally, the performance of existing prediction systems is reviewed, showing that, although more predictive capability remains than what is presently achieved, both the near- and long-term forecasts show skill over trivial predictors. Parallel efforts should therefore be invested into acquiring more observations of the ocean and sea ice, developing new models both in standalone and coupled mode, and improving the data-assimilation techniques.