Artificial intelligence news.
Straight from the original announcements.

AWS brings OpenFold3 biomolecular structure prediction to SageMaker

November 13, 2026, 22:53By the AI News Bites Editorial TeamSource: AWS
AWS brings OpenFold3 biomolecular structure prediction to SageMaker
Image: AWS

AWS says users can deploy OpenFold3-preview through Amazon SageMaker JumpStart to predict 3D structures for proteins, DNA, RNA, small-molecule ligands and the complexes they form. The managed service is designed to handle the GPU infrastructure, request queuing and model serving needed for predictions that can take several minutes.

Key points
  • OpenFold3-preview is available through Amazon SageMaker JumpStart.
  • The model predicts structures for proteins, DNA, RNA, ligands and their complexes.
  • Asynchronous inference supports jobs with invocation timeouts of up to 1 hour.
  • Results and error details are written to Amazon S3.
  • OpenFold3 is released under the Apache 2.0 license as a research preview.

An all-atom model for biomolecular complexes

The OpenFold Consortium and the AlQuraishi Lab at Columbia University developed OpenFold3 as an open source biomolecular structure prediction model. AWS describes OpenFold3 as an all-atom model that predicts the joint 3D structure of molecules from their sequences.

Unlike a model limited to single protein chains, OpenFold3 can predict multi-chain protein complexes, protein-nucleic acid assemblies and protein-ligand complexes in one prediction. Inputs include protein, DNA and RNA sequences, as well as small-molecule ligands supplied as SMILES. Results are provided in mmCIF format with pLDDT, PAE and PDE confidence metrics.

The JumpStart listing identifies the model as OpenFold3 0.4.1 with the of3-p2-155k checkpoint. Its model ID is huggingface-proteinfolding-openfold3, and supported instances include ml.g6.2xlarge, ml.g6.4xlarge, which AWS lists as the default, and ml.g6.12xlarge.

Asynchronous inference is built for long-running jobs

AWS says structure prediction can be difficult to operate because it requires GPU infrastructure, a CUDA and PyTorch stack, large model checkpoints and a serving layer for requests that take minutes rather than milliseconds. OpenFold3 runs in an AWS-maintained container on a managed Amazon SageMaker AI endpoint.

The deployment uses SageMaker Asynchronous Inference rather than a real-time endpoint. Users upload a request to Amazon Simple Storage Service, or Amazon S3, and call InvokeEndpointAsync, which returns immediately. The endpoint queues the request, processes it with an invocation timeout of up to 1 hour, and writes the result or error details to an S3 location chosen by the user. Results can be polled or delivered through Amazon Simple Notification Service.

This setup allows teams to submit batches without building their own job queue. AWS also says asynchronous endpoints can scale to zero instances when the queue is empty, helping avoid charges for idle instances between experiments. Users can run several predictions in parallel on multi-GPU instances and add autoscaling.

Research uses and data controls

AWS lists target and complex structure prediction, protein-ligand co-folding, antibody and protein engineering, and high-throughput screening among the intended use cases. Teams can queue batches and process results directly from Amazon S3. They can also supply their own multiple sequence alignments to improve prediction accuracy while helping keep sequences private.

OpenFold3's developers release its code and model weights under the Apache 2.0 license, and publish the training data and methods. AWS says research organizations and pharmaceutical companies can inspect, fine-tune and build on the model under that license. The developers describe OpenFold3-preview as a research preview.

The endpoint runs with network isolation, which AWS says can help keep sequences within a user's AWS account. AWS also advises considering a dedicated S3 bucket or prefix, server-side encryption with an AWS Key Management Service key, restricted bucket access and lifecycle rules for request and result objects. SageMaker endpoints are billed per instance-hour while running, including idle time, so users must clean up endpoints or configure scale-to-zero to avoid unexpected charges.

Related news

AWS shows how to route coding agent requests with NeMo Switchyard
For developersNovember 13, 2026, 23:30

AWS shows how to route coding agent requests with NeMo Switchyard

AWS has published a sample deployment that routes coding agent requests between models on Amazon Bedrock using NVIDIA NeMo Switchyard and LiteLLM. The approach measures…

NVIDIA Nemotron 3.5 Super VL arrives in AWS SageMaker JumpStart
ModelsNovember 9, 2026, 21:01

NVIDIA Nemotron 3.5 Super VL arrives in AWS SageMaker JumpStart

AWS has made NVIDIA Nemotron 3.5 Super VL available in Amazon SageMaker JumpStart, giving users deployment options for a multimodal model that accepts images, video and…

Meta's Muse Spark Open lands in Amazon SageMaker JumpStart
ModelsOctober 15, 2026, 19:43

Meta's Muse Spark Open lands in Amazon SageMaker JumpStart

Meta's Muse Spark Open, an open-weight model optimized for coding, is now available in Amazon SageMaker JumpStart. Users can deploy it to an endpoint in their own AWS…