The dynamic-thermodynamic Louvain-la-Neuve sea ice model (LIM) has been coupled to the OPA primitive equation ocean general circulation model.In the ORCA2-LIM configuration, the model is run on a global domain with 2$^{\circ}$ degrees mean resolution.Model runs are forced with a combined dataset consisting of daily NCEP/NCAR reanalysis data and various climatologies.The model's performance is evaluated with respect to the representation of sea ice and the high latitude oceans.The annual cycle of sea ice growth and decay is realistically captured in both hemispheres, with ice extent, thickness and drift in close agreement with observations.The location of the main sites of deep convection (Labrador and Greenland Seas; continental shelves of marginal seas of the Southern Ocean) is well reproduced.Model deficiencies include a slight overestimation of summer ice extent in the Arctic, and a significant underestimation of multi-year ice in the Weddell Sea. Furthermore, the width of the Arctic Ocean Boundary Current and the Antarctic Circumpolar Current is overestimated.Sensitivity studies indicate that the use of the combined forcing dataset is crucial to achieve a reasonable summer sea ice coverage and that the direct use of the NCEP/NCAR wind stress data leads to an overestimation of sea ice drift velocities.A restoring of sea surface salinity is necessary to avoid spurious open ocean convection in the Weddell Sea.