Streamflow prediction is critical for Brazil, where over 70% of power generation depends on dams. Traditional hydrological models, such as Soil Moisture Accounting Procedure (SMAP) and Linear Stochastic Model (LSM), coexist without clear superiority. In recent years, machine learning, including Artificial Neural Networks (ANN), Support Vector Machines (SVM), and Long Short-Term Memory (LSTM), has gained prominence. Convolutional Neural Networks (CNN) excel in image recognition, and Recurrent Neural Networks (RNN) show promise in forecasting. Most RNN studies focus on short-term predictions, neglecting climate inputs. Calibrating and validating models remain active research topics, while Global Climate Models (GCMs) and regionalization play pivotal roles in understanding climate change impacts. A comparative study of neural networks and mainstream tools for streamflow prediction in 25 Brazilian basins using precipitation and evapotranspiration inputs provides insights into future water management amid varying climate conditions and demands. Overall, the ANN showed excellent performance, often above 0,9 Nash-Sutcliffe coefficient, on the training dataset, but failed on the validation step due to sample size except in a few basins.