Climate change makes water availability an increasing problem in regions where agriculture is dependent on the seasonality and frequency of rainfall, such as the Rio Santa basin in Ancash, Peru. There, the annual precipitation cycle is characterised by a dry and a rainy season interrupted by dry and wet spells. Climate modelling is a useful tool to quantify future changes in these patterns, provided that the models can be trusted. Therefore, it is necessary to first adjust climate models to a specific domain or use case and test their performance before applying them for projections. This study is the first performance assessment of the COSMO climate model (CCLM) in simulating the climate of the Peruvian Andes at high resolution. ERA5 was used to drive two one-way nested domains with 12 km and 2.2 km grid spacing. First, the sensitivity of the model was evaluated in terms of deep convection parameterization, model top, Rayleigh damping layer height and type. The analysis showed that using a deep, instead of a shallow, convection scheme in the first domain results in more precipitation in most parts of the domain. Using such a scheme also leads to a displacement of the Bolivia high (BH), the main driver of the regional climate. In contrast, no sensitivity to the other parameters is evident for the chosen simulation period of one month. The output of a full-year simulation realised with the most promising setup is compared with satellite precipitation estimates, reanalysis products, in-situ measurements and simulations performed using the Weather Research and Forecasting (WRF) model. The results reveal that CCLM overestimates orographically induced precipitation, leading to a significant wet bias across the Andes. CCLM correctly reproduces the upper-level circulation pattern but underestimates the BH, leading to uncertainties in the transition months, which limits the model’s ability to determine the beginning and end of seasons. However, the dry and wet seasons are well reproduced and CCLM shows satisfying skills in reproducing the temporal distribution of precipitation: the timing of dry and wet spells is accurate, although the wet periods are too intense. Furthermore, CCLM’s representation of the physical processes driving these spells is largely consistent with previous studies. Compared to WRF, CCLM has some discrepancies in the spatial precipitation pattern but reproduces the diurnal cycle better. After a bias correction, CCLM should be able to simulate future precipitation climate in a reasonable way.