Intrinsically disordered proteins (IDPs) undergo sequence-dependent liquid-liquid phase separation forming membraneless organelles or biomolecular condensates. These condensates are crucial in a number of cellular processes such as signaling, transcription and stress response, and also find many uses in the design of biomaterials. An open question in the field of protein phase separation pertains to how the amino acid sequence of the IDP encodes its phase behavior and the possibility of developing a sequence-to-phase behavior relationship for IDPs. This relationship is often referred to as the “molecular language” or “molecular grammar” of protein phase separation. In this work, we investigate condensates formed by a model polypeptide at atomisitic resolution and find a diverse range of interactions that stabilize the condensed phase. Using these insights, we systematically design sequence variants of the model IDP, targeting specific sequence features such as net charge, aromaticity and polarity and perform experiments and simulations to quantify the relative importance of these features in determining the phase behavior. Through these mutants, we find that enthalpically favorable interactions between residue pairs beyond the scope of current sequence heuristics play a significant role in determining the phase behavior of IDPs. To further resolve the roles of individual amino acids in determining the phase behavior of IDPs, we quantify the free energies of transfer of all amino acids from the dilute phase to the dense phase of model condensates. We find that the resulting free energies provide a fundamental understanding of the thermodynamic driving forces underlying phase separation at level of a single amino acid, capturing key experimental trends such as the comparable driving forces imparted by polar and aliphatic amino acids, the high favorability of aromatic amino acids and the apparent preference for positive over negative charge in phase separating IDPs. We further resolve the protein and water-mediated interactions and find a delicate balance between favorable protein-mediated and unfavorable water-mediated interactions in determining the thermodynamic driving forces underlying condensate formation. Taken together, our results provide a thermodynamic framework for interpreting experiments on sequence-dependent phase separation and pave the way for the future development of predictive models for the design of stimulus responsive biomaterials.