Model Hub is a dedicated space inside Benchling where scientists can discover, run, and track scientific AI models. From the Model Hub icon in the left navigation bar, you can browse a curated library of structure prediction and generative models, submit predictions on individual sequences or across hundreds at once, and follow results through to experimental execution.
If Model Hub is not enabled on your tenant, reach out to your Benchling representative or Benchling support at support@benchling.com. Model Hub is automatically available by default for most tenants.
The existing structure prediction experience on AA sequences remains available — Model Hub adds a new, dedicated entry point that is not dependent on working from a specific entity record, and enables bulk runs across large sets of sequences.
Access Model Hub
Model Hub is accessible from a new icon in the Benchling left navigation bar. If you don't see the icon, contact your Benchling representative or Benchling Support to have it enabled.
1. Choose your models
Browse the curated model library and select the right tool for your question. Structure prediction, generative design — the library is organized to help you find what fits your experiment.
2. Choose your data
Select sequences and compounds already registered in your Benchling registry. API support and direct data upload (for data not yet in Benchling) are coming soon.
3. Run
Submit a single prediction or a Prediction Batch across hundreds of sequences. Benchling handles the compute and will notify you when the predictions are complete. View your Prediction Batches or individual Prediction Jobs through their tabs on the Model Hub page.
4. Act on results
Once the status for a prediction is Succeeded, you can download the full outputs directly from the model (in the format the documented model
[Image: Prediction Jobs log showing completed predictions with provenance and ordering options]
Available models
The following models are at general availability in Model Hub:
- AlphaFold 2
- Chai-1
- Boltz-2
- OpenFold 2 — an open, trainable reimplementation of AlphaFold 2, developed by the OpenFold Consortium and the AlQuraishi Lab at Columbia University
- OpenFold 3 — an open source reproduction of AlphaFold 3, also from the OpenFold Consortium
- Protenix — ByteDance Research's structure prediction model for proteins and small molecule ligands as part of a single complex
BoltzGen, a generative protein model, remains in Beta and is not part of the general availability release.
Model Hub is designed to grow continuously with both open source advances and proprietary model partnerships. More model types and modalities — including generative design and broader molecule types — are on the roadmap.
Prediction Batches and Prediction Jobs
Prediction Batches allow you to select a large set of sequences and submit them all at once with the same model configuration. This is particularly useful for teams running structure predictions across large candidate libraries. Results come back as an organized Prediction Batch, ready to review and act on.
Prediction Jobs are the individual predictions that make up a Prediction Batch — each sequence or input submitted in a batch generates its own Prediction Job. You can track the status and results of each prediction independently, rather than only seeing a batch-level summary. From the Prediction Jobs view, you can review outputs for specific inputs, trace results back to source sequences or experiments, and share findings with collaborators.
MSA support
MSA stands for Multiple Sequence Alignment. Some structure prediction models (like OpenFold) use evolutionary context — i.e., how similar proteins look across many species — to produce higher-confidence predictions.
Benchling runs GPU-accelerated MSAs, which are denoted on specific models by the "Use MSAs" checkbox.
Access and administration
- Model Hub is automatically available by default for most tenants. Scientific AI Models (including the Model Hub Page) can be disabled from the Tenant Admin Console.
- Model Hub is not available on Validated Cloud.
- Existing structure prediction runs are not affected when Model Hub is enabled. All prior results remain saved and viewable, including older predictions.