OpenFold3-preview now deployable via Amazon SageMaker JumpStart

OpenFold3 is an all-atom structure prediction model that folds multi-chain protein complexes, protein–nucleic acid assemblies, and protein–ligand complexes in a single prediction. The developers describe OpenFold3-preview as a research preview, and it runs in an AWS-maintained container on a managed Amazon SageMaker AI endpoint. The endpoint uses asynchronous inference because folding time grows with the number and length of chains in a query, and real-world proteins often take several minutes, exceeding the 60-second cap of a real-time endpoint. This design queues bursts of requests internally and can scale to zero instances when the queue is empty, which helps avoid paying for idle instances between experiments. For proprietary sequences, the endpoint runs with network isolation and accepts precomputed MSAs, so teams generate alignments on infrastructure they control rather than sending sequences to a public server. In a test co-folding carbonic anhydrase II with zinc and acetazolamide, supplying an MSA substantially improved confidence metrics and recovered the known binding mode, while prediction time stayed around 1.5 minutes.