AlphaFold 3

AlphaFold 3

AlphaFold 3 is a biomolecular structure-prediction model described by researchers from Google DeepMind and Isomorphic Labs in a Nature paper published on May 8, 2024.

What it predicts

The model predicts the joint structures of complexes that can include proteins, nucleic acids, small molecules, ions, and modified residues. The paper introduced a diffusion-based module for generating atomic coordinates.

This is broader than predicting the folded structure of a protein alone. It allows the model to address several kinds of interactions within one system.

What the paper reported

The authors evaluated AlphaFold 3 across protein-ligand, protein-nucleic-acid, antibody-antigen, and other structure-prediction categories. They reported higher accuracy than the compared specialized methods in most of those categories.

Those results belong to the named datasets, metrics, comparators, and evaluation protocol. They do not establish that AlphaFold 3 replaces every physics-based method or laboratory experiment.

The paper also discusses limitations, including hallucinated structure in some diffusion outputs and the role of confidence measures. A predicted structure remains a model output that must be interpreted in its scientific context.

Access boundary

At launch, AlphaFold Server offered free access for non-commercial research. Google DeepMind later made AlphaFold 3 model code and weights available for academic use. "Free to use through a server" and "open source" are not interchangeable descriptions.

Access terms, model versions, supported inputs, server limits, licenses, and research guidance require a current check.

How E016 used it

E016 paired AlphaFold 3 with GPT-4o because both launches suggested a reduction in friction. The canonical article narrows that comparison. AlphaFold 3 can reduce friction in forming structural hypotheses, while scientific value still depends on validation beyond the prediction.

Editorial and verification notes

The profile was checked against the original Nature paper, Google DeepMind's May 2024 launch material, and its current AlphaFold page on July 25, 2026. It is not a laboratory protocol or a claim of clinical utility.