Description: The Arctic Ocean is a challenge to model accurately. Its exchanges with the rest of the global ocean occur through narrow gateways. Ventilation within the Arctic requires a realistic continental shelf hydrography and slope, interaction with the sea ice and atmosphere, and preservation of dense overflows. At all depth levels, an accurate bathymetry is needed to properly represent the circulation. The uppermost layers depend on both surface heat fluxes and freshwater fluxes from rivers, glaciers, sea ice, and the atmosphere, while the deepest layers are impacted by geothermal heating. Despite this, many parameterisations and tuning processes applied in the Arctic are not representative of the polar regions. In addition, observations used to constrain Arctic models are often limited to the summer season, ice-free regions, or upper ocean. Therefore, unsurprisingly, the coarse-resolution CMIP-type models are highly inaccurate in the Arctic Ocean. In this presentation, we review a non-exhaustive list of biases in Arctic Ocean water mass representation and circulation in CMIP6 models, with a specific focus on how these biases impact our ability to accurately project future Arctic Ocean and global changes. Key directions for improving the Arctic Ocean in climate models will be discussed. We will finish with promising examples of ongoing Arctic model development: regional high-resolution modelling to improve simulations of sea ice and the exchanges with the global ocean, including overflow representation; pan-Arctic modelling with an adaptative mesh to improve the representation of mesoscale eddies and mixing; and the nudging of an ultra-high-resolution model (250 m) against observations.