COSMO-NET is a deep learning framework that revolutionizes the prediction of COSMO-SAC sigma profiles and thermodynamic properties directly from molecular structures. Built on state-of-the-art graph neural network architectures—including Directed Message Passing Neural Networks (D-MPNN) and Graph Convolutional Networks (GCN)—COSMO-NET eliminates the need for computationally expensive quantum chemistry calculations by learning from a curated database of more than 16,000 compounds, enabling rapid and accurate prediction of COSMO-based molecular descriptors for applications in solvent selection, process design, and drug discovery.