This study aims to optimize mix designs of geopolymer pervious concrete using multi-phase discrete element modeling. Metakaolin based geopolymer concrete was prepared with different mix designs for laboratory testing. A multi-phase discrete element model (DEM) for pervious concrete was developed with randomly generated porous structure and realistic contact models between each component. Contact model parameters were calibrated using orthogonal analysis based on measurements of mechanical strengths and elastic modulus. Results of model validation indicate that the developed numerical model has good reliability to predict mechanical properties of pervious concrete with the influence of porosity. Both actual and simulated pervious concrete specimens present consistent crack propagation patterns and failure behavior. The optimization of mixture design is conducted by considering aggregate gradation and aggregate to paste ratio of geopolymer pervious concrete, which are based on experimental results and post-calibration simulations. The gradation with larger size aggregate is suitable for preparing geopolymer pervious concrete with higher permeability and tensile strength. More importantly, it is feasible to reduce paste content in geopolymer pervious concrete while maintaining similar or even higher mechanical strengths, which reduces the embedded carbon footprint. The study findings demonstrate the beneficial use of numerical modeling to determine the optimized mix design of composite material such as geopolymer pervious concrete.