Solubility of drugs is a key piece of information for the pharmaceutical industry. Despite its importance, particularly at the beginning of a new drug process development, this thermodynamic property of the solid-liquid equilibria (SLE) can hardly be predicted for a given molecule in a given solvent In our recent works, some thermodynamic models (UNIFAC and its modifications, COSMO-SAC and its refinements, NRTL-SAC) were investigated and compared for solubility prediction. The main drawbacks of these methods concern the strongest molecular interactions (dipole-dipole, hydrogen bonding), which are not properly taken into account. In the present study, we propose a new optimization of the last two COSMO-SAC refinements (2007 and 2010) for solubility prediction. To do that, a parameters optimization upon 352 solubility data of complex organic molecules was performed. Also, to improve hydrogen bonding influence, new a-profiles are generated by applying another probability function for the solute. The results of this work are encouraging, especially for the calculation of crystallization yields given that the solubility temperature dependence is well represented. Solubility predictions in polar solvents are improved. However, this improvement is not as good as expected since it seems that the parameters balance between hydrogen bonding and electrostatic interactions is not the best. To confirm this, a short computation of anthracene solubility in toluene and heptane was performed.