AlphaFold 3
Diffusion-based AlphaFold 3 predicts structures of proteins, DNA, RNA, ligands and ions together, with 50%+ better protein-ligand accuracy than prior methods.
Replaces AF2's structure module with a diffusion network on an improved Evoformer. First AI system to beat physics-based tools on PoseBusters (company claim). Free AlphaFold Server for non-commercial use; code and weights for academics released 2024-11.
- Date
- Wednesday, 8 May 2024
- Lab
- Google DeepMind
- Kind
- model
- Access
- open weights (restricted license)
Figures
| Measure | Value | Measured by |
|---|---|---|
| Protein-ligand interaction accuracy | 50%+ better vs existing methods; doubled for some categories (company claim) | company |
Published in Nature 2024-05-08 with Isomorphic Labs. Launch offered AlphaFold Server only; code and weights for academic use followed in 2024-11. A preview of the next-gen model appeared 2023-10-31.
Sources
- blog.google/technology/ai/google-deepmind-isomorphic-alphafold-3-ai-model
- deepmind.google/blog/a-glimpse-of-the-next-generation-of-alphafold/
This record was checked against its sources on 6 October 2026. How we check