Boltz Bio launches AI platform for protein design and molecule screening

Boltz Bio moved its AI work from open model releases into a more complete discovery stack, pairing new protein-design and molecule-screening models with an API and deployable workflows. The company said the release centers on BoltzProt-1 for binder design and BoltzMol-1 for small-molecule work, with immediate access through a partnership with Tamarind Bio.

The clearest signal in the launch is not a benchmark score, but what Boltz says researchers can now do more reliably. Across 10 novel targets, BoltzProt-1 raised confirmed-binder hit rate from 3.3% to 8.0%. In a separate test set drawn from prior literature, the model produced nanobody screening hits on 7 of 10 additional targets. Boltz also said 58% of confirmed binders met all reported developability criteria, compared with 40% for BoltzGen and 21% for clinical-stage VHH controls.

That matters because discovery teams do not live on leaderboard rankings alone. BoltzProt-1 is positioned as a de novo binder-design pipeline for protein binders and nanobodies, while BoltzMol-1 expands the company’s reach into small-molecule design and screening. Boltz says its API can predict structure and binding, design proteins and small molecules, and screen libraries, with confidence and binding metrics returned as part of the workflow. The platform is offered with multi-tenant, single-tenant, and on-premise deployment options, alongside SOC 2 Type 1 verification.

Boltz is also leaning hard into scale. The company says its models are used by more than 100,000 scientists, and its website now describes the business as a frontier AI lab serving biopharma, agriculture, academia, and consumer products. BoltzLab documentation says protein-binder design can run as virtual screens across thousands of candidates, with guidance to start with 30 to 100 designs before expanding campaigns to 10,000 to 100,000 designs.

The new release builds on an earlier arc that helped define Boltz’s reputation. Boltz-1 was described by the company as the first fully open-source model to approach AlphaFold3 accuracy for biomolecular structure prediction, and Boltz-2 extended that foundation by jointly modeling structure and binding affinity. BoltzGen, the earlier generative model, is described by Boltz as an all-atom diffusion model for designing proteins and peptides against biomolecular targets, and the company says it was validated in a large-scale distributed effort with academic and industry labs. The new product layer, backed by the January 8, 2026, $28 million seed round led by Amplify, a16z, and Zetta, shows Boltz pushing from model novelty toward infrastructure for discovery.