Multi-fluid mixture models for highly accurate multiparameter equations of state have been applied very successfully in the past years in order to accurately model thermophysical properties of mixtures. The multi-fluid mixture model mainly relies on empirical reducing functions and for some mixtures also on departure functions, for which the mathematical structure is a priori unknown and thus must be determined during the fitting procedure. By applying standard mixing rules for the reducing functions and by omitting the departure function, the mixture model can also be used predictively. However, it is demonstrated in this work that the predictive capability of this type of mixture model is rather limited. Therefore, a new model is proposed, which is a combination of the multi-fluid model with excess Gibbs energy models. This new approach results in a theoretically based formulation for the departure function of the multi-fluid model. It is shown that the new model yields very good results for the description of binary mixtures of the components ethanol, ethane, carbon dioxide, propene, and benzene. While the state-of-the-art multi-fluid model with either predictive linear mixing rules or Lorentz-Berthelot combining rules for the parameters of the reducing functions does not represent the phase equilibria for the investigated binary mixtures well, and in case of the azeotropes predicts qualitatively wrong mixture behavior, the new model is capable of accurately representing the phase equilibria of all binary mixtures investigated.