AlphaFold 3 achieves 50% accuracy gain on protein-ligand binding over prior methods, available free via server for non-commercial research

Google DeepMind and Isomorphic Labs released AlphaFold 3, a biomolecular structure prediction model that claims 50% improvement in protein-molecule interaction prediction versus existing methods, and first AI system to surpass physics-based tools on binding-site benchmarks. The model handles proteins, DNA, RNA, ligands, and chemical modifications in a single end-to-end prediction. AlphaFold Server was launched free for non-commercial academic research. The advance expands molecular design capabilities beyond proteins to a broad spectrum of biomolecules. For financial institutions: this is a new frontier application claim backed by benchmarks, signaling expansion of AI advantage in R&D workflows but without disclosed commercial revenue or enterprise ROI data.