AWS has published a walkthrough for deploying OpenFold3-preview on SageMaker JumpStart. The November 13 post describes a managed, asynchronous endpoint for predicting structures of proteins, DNA, RNA, small molecules and combinations of them.
This is a new deployment option for an existing research-preview model, not a claim that OpenFold3 itself was newly released. The model was developed by the OpenFold Consortium and Columbia University’s AlQuraishi Lab. AWS specifies version 0.4.1 and the of3-p2-155k checkpoint in its example; the upstream project now documents newer parameter options and marks this checkpoint deprecated. Its parameter reference says this older checkpoint requires a compatible OpenFold3 version.
Why asynchronous inference
Structure predictions can take longer than a short web request, especially as the number and length of input chains grow. AWS describes placing a request in object storage, invoking an asynchronous endpoint and retrieving results or errors from storage afterward. Its example supports selected GPU instance types and describes network isolation for the endpoint.
The published workflow uses paid AWS infrastructure, and data storage, GPU time and other services may incur charges. Organizations should review actual pricing, data handling, encryption and retention before deploying sensitive sequences.
Predictions are hypotheses
OpenFold3 predicts molecular structures; a plausible predicted structure does not establish a drug’s efficacy, a treatment’s safety or a verified biological interaction. Confidence measures can help researchers interpret outputs but do not replace experimental validation. AWS lists Apache 2.0 for the cited preview; the repository profile and model profile link to the upstream documentation and current terms. The Hugging Face repository requires users to accept access conditions.
We have not deployed the endpoint or independently evaluated the model.
Original reporting; primary sources reviewed November 14, 2026. No independent hands-on test claimed.
