AlphaFold 3 is Google DeepMind’s landmark model for predicting the structure and interactions of life’s molecules — proteins, DNA, RNA, ligands and more — and this repository is its official inference pipeline. Released under the Apache-2.0 license for the code, it has collected more than 8,600 stars on GitHub and was actively maintained through 2026.
AlphaFold 2 transformed structural biology by predicting protein shapes; AlphaFold 3 generalizes the idea to joint structures of biomolecular complexes, including how proteins bind small molecules and nucleic acids — the interactions that matter most for drug discovery. This repo contains the full open inference code: feed it a JSON description of your sequences and ligands, and it runs the diffusion-based model to produce predicted structures with confidence metrics. Model weights are distributed separately by DeepMind on request for academic and non-commercial research use. The release accompanies the AlphaFold Server for casual use, but this self-hosted pipeline is what research groups run when they need batch predictions, custom inputs or results that never leave their own infrastructure — a workflow that has made it a standard component of computational-biology toolchains worldwide.
The inference code is free and open source under the Apache-2.0 license. Model weights are provided free of charge on request for non-commercial research use under DeepMind’s terms.
Since its release, the pipeline has been widely adopted by academic core facilities and biotechnology groups, and its predictions feed downstream tools for docking studies, protein design and variant-effect analysis.
It is for structural biologists, computational chemists and drug-discovery researchers who need state-of-the-art biomolecular structure prediction running on their own infrastructure.