In carbon therapy, the interaction of the incoming beam with human tissue may lead to the production of a large amount of nuclear fragments and secondary light particles. An accurate estimation of the biological dose on the tumor and the surrounding healthy tissue thus requires sophisticated simulation tools based on nuclear reaction models. The validity of such models requires intensive comparisons with as many sets of experimental data as possible. Up to now, a rather limited set of double differential carbon fragmentation cross sections has been measured in the energy range used in hadron therapy (up to 400 MeV/nucleon). However, new data have been recently obtained at intermediate energy (95 MeV/nucleon). The aim of this work is to compare the reaction models embedded in the geant4 Monte Carlo toolkit with these new data. The strengths and weaknesses of each tested model, i.e., G4BinaryLightIonReaction, G4QMDReaction, and INCL++, coupled to two different de-excitation models, i.e., the generalized evaporation model and the Fermi break-up model, are discussed.