Accurate solar irradiance forecasting is pivotal for the effective integration of photovoltaic systems into modern power grids. Recent advances in artificial intelligence have led to more advanced models capable of capturing the highly nonlinear and time-dependent patterns. In this work, we compare the performance of several state of the art and emerging AI architectures for short-term time series predictions in the context of global horizontal irradiance forecasting. A comprehensive dataset comprising historical ground-based irradiance and weather measurements is employed to train and validate the models. Evaluation metrics, such as mean absolute error and root mean square error, are utilized for models’ performance evaluation. Furthermore, an assessment of the model’s robustness under varying meteorological conditions will be conducted. The findings provide a benchmark for comparing these models, depicting their respective strengths and limitations subsequently advancing the data-driven solar irradiance forecasting.