The predictive capability of gas and liquid solubility in organic polymers is examined based on the combination of the PR+COSMOSAC equation of state (EOS) and the COSMO-SAC liquid model through three different excess Gibbs free energy based mixing rules, modified Huron-Vidal (MHV1), Wong-Sandler (WS), and self-consistent mixing rule (SCMR). Using 81 binary systems consisting of 23 gas molecules and 22 polymers (81 data points) with temperatures ranging from 298 to 461 K, it is found that WS and SCMR can provide reasonable prediction accuracy (RMSE(log10kH) = 0.746 and 1.725, respectively) for the Henry's law parameter in polymers, while the MHV1 mixing rule results in a much larger error (RMSE (log10 kH) = 3.118) compared to experiment. The WS and SCMR, but not MHV1, provide a converged value of Henry's law parameter of gas in polymers as the molecular weight of the polymer increases. We further propose a modification to the SCMR (mSCMR) that results in significant improvement in the solubility prediction in polymers (RMSE (log10 kH) = 0.305) and the binary vapor-liquid equilibrium for common molecules. In this new approach, referred to as PRCS/mSCMR/COSMOSAC, all species-dependent parameters are determined from quantum mechanical (QM) calculations, and no adjustable parameters are required for the gas-polymer binary pairs. We believe that this new method may provide useful assistance to the development of polymer membrane-based gas separation processes especially when experimental information is not yet available.