Although strong binding to the target protein is a prerequisite, it is not enough to be an effective drug. To produce a particular functional response, drugs need to regulate the targets’ signal transduction pathways, either blocking the proteins’ functions or modulating their activities by changing the conformational equilibrium. The routinely calculated binding free energy of a compound to its target is a good predictor of affinity but may not always predict efficacy. While the time scales for the protein conformational changes are prohibitively long to be routinely modeled via physics-based simulations, thermodynamic principles suggest that binding free energies of the ligands with different receptor conformations may infer their efficacy if the functional response of the receptor is determined by thermodynamics. However, while this hypothesis was proposed in the past, it has not been thoroughly validated and is seldom used in practice for ligand efficacy prediction. We present an actionable protocol and a comprehensive validation study to show that binding thermodynamics provides indeed a strong predictor for the efficacy of a ligand. We apply the absolute-binding free energy perturbation (ABFEP) method to ligands bound to active and inactive states of eight G protein–coupled receptors (GPCRs) and a nuclear receptor. By comparing the resulting binding free energies, we can determine with a very high accuracy whether a ligand acts as an agonist or an antagonist. We find that carefully designed restraints are often necessary to efficiently model the corresponding conformational ensembles for each state and provide a procedure for setting up these restraints. Our method achieves excellent performance in classifying ligands as agonists or antagonists across the various investigated receptors, all of which are important drug targets.