# AWS adds OpenFold3 research preview to SageMaker JumpStart

> AWS users can deploy the open-source structure-prediction model on managed SageMaker endpoints, which process jobs asynchronously and can scale to zero when idle.

Oossa · 2026-11-13 · https://oossa.com/en/aws-adds-openfold3-research-preview-to-sagemaker-jumpstart

AWS has made OpenFold3-preview available through Amazon SageMaker JumpStart, its model deployment service. Developed by the OpenFold Consortium and Columbia University’s AlQuraishi Lab, OpenFold3 predicts structures for proteins, DNA, RNA, small molecules and the complexes they form.

The endpoint handles predictions that can take minutes by queuing jobs and returning results through Amazon S3. AWS says users can configure it to scale to zero between jobs, but running SageMaker instances are billed by the hour. The model is a research preview, and users need an AWS account and supported GPU capacity.

## The facts

- JumpStart lists OpenFold3 version 0.4.1, checkpoint of3-p2-155k, under the Apache 2.0 license.
- Supported SageMaker instances include ml.g6.2xlarge, ml.g6.4xlarge and ml.g6.12xlarge.

## Why it matters

For research teams with AWS accounts, JumpStart offers a managed way to run biomolecular structure predictions, though GPU instance charges and research-preview status remain practical constraints.

## Sources & references

1. [OpenFold3 on Amazon SageMaker JumpStart: Predict biomolecular structures](https://aws.amazon.com/blogs/machine-learning/openfold3-on-amazon-sagemaker-jumpstart-predict-biomolecular-structures/) – AWS, 2026-11-13

Last updated: 2026-11-13