COSMO-SAC model was reparameterized with use of the critically evaluated data generated by the NIST ThermoData Engine for vapor liquid equilibria, excess enthalpies for binary mixtures, and activity coefficients of binary mixture components. The calculated a-profile library contained 897 individual compounds. The temperature-dependent a profiles included contributions of up to 40 conformers of a molecule. Splitting of the H-bonding a profiles into OH and non-OH parts decreased the root-mean square deviation from the experimental data points by about 10% compared to the model using one H-bonding parameter. The original UNIFAC model demonstrated comparable performance with the more advanced COSMO-SAC variation. The challenges of uncertainty evaluation for parameters of the model and the predicted values are discussed.