The occupant behavior plays a fundamental role in the actual performance of the buildings, especially for very low energy buildings. Since inconsistencies are discovered post-occupancy, known as Energy Performance Gap, it is necessary to better represent the occupants when optimizing the design of buildings. In this context, a multi-objective optimization of a low energy building was carried out for three different French climates with two models of occupant behaviors. The first is a standard fixed scenario (SF) while the second is an agent-based tool (NoMASS) that predicts occupancy scenarios according to sociodemographic characteristics and estimates occupants’ interactions with windows, external shadings and lighting based on probabilistic models. The results obtained show that the occupant model significantly influences the expected performance of the building as NoMASS model estimates greater heating needs and improved summer comfort. However, for our case study, there is no influence of the accuracy of representation of occupant behavior on the optimal solutions of building design. The optimization could be carried out with a less detailed model, thus reducing calculation times while maintaining comparable optimal results. The performance of the optimal solutions obtained can then be refined by simulating with the most detailed model.