Modeling protein-ligand binding is challenging, with methods ranging from physics-based approaches to deep learning. We introduce Protein-Ligand Equivariant Transformer (ProLET), a model based on SE(3) equivariant geometric deep learning, which outperforms existing methods in binding affinity prediction and pose estimation, excelling on numerous benchmarks. ProLET stands as a powerful and adaptive resource, addressing critical stages of drug discovery including lead optimization and hit identification. Our approach marks a step forward towards targeted and accelerated development of therapeutics.