
OpenFold3
A community, bitwise-faithful open reproduction of AlphaFold 3
What's new
OpenFold3-preview is public for community feedback — notably the only open model to match AlphaFold 3 on monomeric RNA structures.
OpenFold3 is the next generation of the OpenFold project, aiming to be a bitwise-faithful, fully trainable reproduction of AlphaFold 3. It is developed by the AlQuraishi Lab at Columbia University together with the OpenFold Consortium.
Highlights
- Apache-2.0 and free for academic and commercial use, with training data released.
- Competitive with the state of the art in open structure prediction.
- The only open model reported to match AlphaFold 3 on monomeric RNA structures.
The project is a research preview; the final model and training pipeline are still in active development.
Related tools

AI Models
AlphaFold 3
Predict the joint structure of proteins, nucleic acids and ligands

AI Models
Boltz-2
Open-source structure and binding-affinity prediction, MIT licensed

AI Models
Protenix
High-accuracy open-source AlphaFold3 reproduction from ByteDance