The PRADA experiment aims at improving the performances of high speed parallel robots in order to address new market shares when accuracy along complex paths is needed. The first step has consisted in deriving efficient dynamics models that can be computed in real-time while being accurate enough. The models for the two versions of the robot (redundant and non-redundant) will be validated. In the case of the non-redundant robot, the derivations exhibit an interesting feature; thanks to its symmetric design, the generalized mass matrix has a very simple expression.