Despite the multisectoral importance of water resources in the context of climate change, it remains still difficult to correctly simulate the terrestrial water cycle via general circulation and Earth system models. While existing model evaluation efforts from the Climate Model Intercomparison Project 5 (CMIP5) are mainly focused on atmospheric variables, we considered here the simulated freshwater flux into the ocean for the river Congo by 20 CMIP5 models. We found that many advancements are still required in global modeling to satisfactorily reproduce the discharge, at least for Congo River. Only the distribution of intermonthly and interannual anomalies of simulated flow, calculated as percent deviations from the respective long‐term average, appears not significantly different from that found in observations. Conversely, most of the models appear to largely overestimate the streamflow seasonal cycle, up to 4–5 times for some months. Further, when looking at severer anomalies in terms of abnormal low‐flow periods (hydrological droughts), the models seem to underestimate their frequency and magnitude while overestimating the duration of deficit conditions with respect to hypothetical water demand. These overall discrepancies between models and observations could propagate in reproducing the land‐ocean feedbacks, as discharge is known to affect sea surface salinity and temperature and, thus, the regional to global ocean circulation and climate. Moreover, biased model‐based discharge simulations could wrongly support water management and infrastructure planning for large river basins, especially when using models instead of measurements, as in case of ungauged rivers, or however when clarification of models' uncertainty is missed.