The range of the predictive Gibbs energy of solvation model, COSMO-SAC, is extended to large ranges of density, pressure, and temperature for very nonideal mixtures by combining it with an equation of state (EOS) using the Wong-Sandler mixing rule. The accuracy of isothermal vapor-liquid equilibria (VLE) calculations based on using the predictive COSMO-SAC model and separately the correlative NRTL model is compared, each combined with three different forms of the Peng-Robinson equation of state. All the models considered require the value of the EOS mixing rule binary parameter k(ij). The NRTL model also requires three other parameters obtained from correlation low pressure VLE data. The PRSV + COSMO-SAC model is showed, with its one adjustable parameter obtained from low temperature data leads good predictions at much higher temperatures and pressures. (C) 2017 American Institute of Chemical Engineers