The Black Sea is one of the largest land-locked basins in the world. Due to the vulnerability of its unique marine ecosystem, accurate long-term modelling of its hydrodynamics is needed. In this study, we first compare the skills of four NEMO based Black Sea models in a free-run which use different discretization schemes. We find that the most accurate results are obtained with the model (named CUR-MEs) which has a 3D mesh optimized for the prevailing dynamics. This new model uses a curvilinear horizontal grid with increased resolution (≈950m) over the shelf-break and lower resolution (≈6km) in areas where the scale of relevant processes is larger (≈20km). In the vertical, CUR-MEs uses Multi-Envelope curved s-levels designed to optimize the representation of the Cold Intermediate Layer (CIL). Second, we compare CUR-MEs in free-run with the data-assimilative CMEMS reanalysis. Validation against independent observations shows that the two models have similar skills - e.g., the difference between the mean BIAS and RMSE of the two models is ≈0.15°C for temperature and ≈0.07 for salinity. The CUR-MEs model, even without data assimilation, is able to correctly reproduce the details of the variability of the Mean Kinetic Energy and the CIL.