Chai Discovery

#Overview

Chai Discovery is the first-party lab behind Chai-1, an AlphaFold3-class model for biomolecular structure prediction. As the model's authors, it is the authoritative place to run Chai-1 for protein and complex structure prediction, whether you want a hosted endpoint or the underlying open weights. For researchers comparing structure predictors or looking for an AlphaFold 3 alternative they can both call as an API and run themselves, Chai Discovery is the canonical source.

#What you can run on Chai Discovery

Chai-1 predicts the three-dimensional structure of proteins and biomolecular complexes, including protein–protein and protein–ligand assemblies, at a capability level comparable to AlphaFold 3. It handles single chains and multi-component systems, making it suitable for modeling interactions and assemblies rather than isolated monomers alone. Because it comes directly from the developing lab, you get the reference implementation and the model behavior the papers describe, without a third-party reimplementation in between.

#Accessing Chai-1

There are two paths. The hosted route runs Chai-1 through a web application and an API, so you can submit structure prediction jobs interactively or programmatically without managing your own GPUs — a good fit for teams that want reliable prediction on demand. The self-host route uses the open Chai-1 weights and the accompanying chai-lab code, letting you run predictions inside your own environment for data-sensitive work or high-throughput pipelines. Newer models from the lab are offered through the commercially licensed hosted interface rather than as open releases, so the open weights apply specifically to Chai-1 while later capabilities are reached through the managed service.