An approach to overcome the limited time and spatial scales in molecular dynamics (MD) simulations is to reduce the number of degrees of freedom in the system through coarse-graining. In hybrid atomistic/coarse-grained (AT/CG) simulations, the region of interest (e.g., the solute) is simulated at a higher level of resolution than the surrounding solvent. Recently, we have reparametrized the interactions between the GROMOS 54A7 force field and the GROMOS CG water model to correctly reproduce solvation free energies of side-chain analogues. In this study, a benchmarking of the AT-CG parametrization using a broad set of 22 proteins is conducted. On the basis of the results, the idea of introducing a thin AT water layer around the solute (protein) is revisited to provide a detailed first solvation shell. Different layer schemes were investigated: (a) the water molecules in the AT water layer around the whole protein were restrained to their respective closest protein atom, and (b) the AT water layer was present only around the charged side chains. Results show that the second scheme is more robust in practice and reduces artifacts from the hybrid model with only a small additional computational cost. In general, this layer scheme was found to preserve the structural properties of the proteins better compared to direct solvation in CG water.