Chai-1

Chai-1 is a multimodal biomolecular structure prediction model from Chai Discovery, released with weights available for non-commercial use and positioned as a practical alternative to AlphaFold 3 for drug-discovery work.

What it does

The model predicts proteins, nucleic acids and small-molecule complexes, and accepts experimental restraints as input — which matters in practice, because a team with partial structural data can fold that knowledge into the prediction instead of discarding it. It is usable through a web interface or run locally against released weights.

Strengths

• Weights released for non-commercial use, enabling local deployment.
• Accepts experimental restraints to condition predictions.
• Competitive accuracy on drug-discovery-relevant targets.
• Backed by a team focused specifically on therapeutic applications.

Limitations

Commercial use requires a licence conversation, so the open-for-research framing has limits. As with every predictor in this class, confidence metrics are guidance rather than proof, and independent benchmarking across diverse targets is thinner than for AlphaFold.