Chai-1 is an AlphaFold 3–inspired model from Chai Discovery, built for predicting the structure of multi-molecule complexes — proteins, ligands, nucleic acids, and their interactions. It performs on par with AlphaFold 3 in independent benchmarks and adds support for user-defined pocket, contact, and docking constraints, making it useful for teams running structure-guided design experiments.

Chai-1 can operate with or without MSAs, enabling faster turnaround in workflows where evolutionary data isn't available or required.

Key capabilities:

• Protein-protein, protein-ligand, protein-nucleic acid complex prediction

• Experimental restraint support (pocket, contact, docking constraints)

• Runs with or without multiple sequence alignments

• Competitive benchmark performance across multimer structure prediction tasks