Extreme rainfall and associated river floodings are important concerns in modern human societies. Despite a recent increase of extreme rainfall, there is no evidence of an increase of the intensity and frequency of flash floods in Southern Europe, and future projections of flash-floods are quite uncertain in part due to the coarse resolution of available climate model simulations.The recent development of convection-permitting climate models allow a better representation of precipitation extremes. This new generation of climate models have been little employed in combination with hydrological models up to now, and their added value for flash flood modeling remains to be identified.In this work, a 2.5-km convection-permitting climate model (CNRM-AROME) simulation is used to force two hydrological models (CREST and GR5H). This new modeling chain is tested in a French mediterranean catchment, the Gardon at Anduze, that experienced severe flash flooding episodes over the last decades. Hydrological models are calibrated using the COMEPHORE 1 km observed precipitation dataset merging radar and rain gauge rainfall at the hourly time step. We compare the CNRM-AROME-based hydrological simulation to a benchmark run driven by a conventional CORDEX 12-km CNRM-ALADIN simulation. The analysis of the peak discharges simulated by both hydrological models driven by the different meteorological inputs allows to determine how higher resolution precipitation could improve the simulation of flash floods.