Fracture modelling in quasi-brittle materials such as rock or concrete is quite complex. For this type of materials, a numerical model must incorporate tools capable of representing the formation and propagation of cracks, and their effect on the structure's global behaviour. Nowadays it is possible to predict, evaluate and understand cracking phenomena in quasi-brittle materials through numerical models, among them detailed rigid particle models (PM) that take directly into consideration the physical mechanisms and the influence of the material grain structure have gained relevance. In this work a 3D rigid PM model is adopted that includes in an approximate way the polyhedral particle shape but keeps the simplicity and the reduced computational costs of rigid spherical particle models [1]. Three constitutive contact models that include softening are briefly described, namely an approximate vectorial bilinear weakening model (BL) [2], a microplane type model that has been proposed within the framework of particle models with a lower degree of particle discretization (MCP) [3], and an elasto-plastic interface model that has been proposed for zero-thickness interface elements (EP) [4]. The performance of the constitutive contact models is initially compared in simple tensile, pure shear and compression/shear tests of a single contact. Additionally, compression, direct tensile and confined triaxial tests of quasi-britlle material discretized with spherical particles are presented and the predicted macroscopic response of the different constitutive models is compared. It is shown that the macroscopic response predicted with the three different contact models can be slightly different from each other for similar contact properties, whereas a very similar behaviour is predicted for a single contact model. The way the contact model handles the damage evolution under tensile/shear or compression/shear strongly influences the overall macroscopic response. Under compression, the MCP contact model tends to predict a less brittle behaviour when compared with other contact models and for rock fracture it may require a particle assembly with a lower coordination number. The BL contact model is shown to be more suitable to use in detailed PM models due to the lower associated computational costs and to the lower number of associated contact parameters that need to be calibrated.