The performance of different Quantum Mechanics/Molecular Mechanics embedding models to compute vacuo-to-water solvatochromic shifts are investigated. In particular, both non-polarizable and polarizable approaches are analyzed and computed results as compared to reference experimental data. We show that none of the approaches outperforms the others and that errors strongly depend on the nature of the molecular transition. Thus, we prove that the best choice of embedding model highly depends on the molecular system and that the use a specific approach as a black-box can lead to errors and sometimes totally wrong predictions.