Boltz

Boltz is an open-source biomolecular structure prediction model developed at MIT, and the most credible fully open alternative to AlphaFold 3. It predicts structures of proteins, nucleic acids and their complexes with small molecules, and unlike the AlphaFold 3 server it can be run locally under a permissive licence.

What it does

The model handles complexes rather than single chains, which is where most of the biologically interesting questions live — protein-ligand, protein-nucleic acid, and multi-chain assemblies. Weights and code are released openly, so it can be deployed on your own hardware, fine-tuned, and built into automated pipelines without an access agreement.

Strengths

• Genuinely open: weights and code under a permissive licence.
• Runs locally, so structures for confidential targets never leave your infrastructure.
• Handles complexes, not just single-chain folding.
• Composable into automated pipelines and downstream tooling.

Limitations

Local inference needs real GPU capacity. Accuracy is competitive but not uniformly equal to the best closed models on every target class, and confidence metrics deserve the same scepticism you would apply to any predictor — a high score is not an experimental structure. See our coverage of the Boltz full-stack release.