Abstract. The sensitivity of a shelf sea model of the Gulf area to changes in the bathymetry, lateral and vertical resolution, vertical coordinates and river and atmospheric forcing is explored. Two new Gulf models with a resolution of 1.8 km, named GULF18-3.6 and GULF18-4.0, differing only in the vertical coordinate system and the NEMO code base employed (NEMO-3.6 and NEMO-4.0.4, respectively) are introduced. We compare them against the existing 4 km PGM4 model, which is based on NEMO-3.4 and is developed and used by the Met Office. PGM4 and GULF18-3.6 use similar types of quasi-terrain-following vertical levels, while GULF18-4.0 employs the multi-envelope method to discretise the model domain in the vertical direction. Our assessment compares non-assimilative hindcast simulations of the three Gulf models for the period 2014–2017 against available observations of tides, hydrography and surface currents. Numerical results indicate that both high-resolution models have higher skill than PGM4 in representing the sea surface temperature and the water column stratification on the shelf. In addition, in the proximity of the shelf break and the deep part of the domain, GULF18-4.0 generally presents the highest accuracy, demonstrating the benefit of optimising the vertical grid for the leading physical processes. For the surface currents, the three models give comparable results on the shelf, while the higher-resolution models might be prone to the double-penalty effect in deeper areas. For the tides, PGM4 has a better skill than GULF18 models, and our tidal harmonic analysis suggests that future work may be needed in order to get real benefit from using a more realistic bottom topography, as in the case of the GULF18 models.