Boltz-2: structure and affinity in one pass
3D protein structure and binding affinity — MIT license, no queue
- Video memory
- 24 GB
- Setup
- ~15 min
- Access
- port 8888
- Billing
- hourly, in BRL
Boltz-2 predicts the 3D structure of proteins, DNA, RNA and ligands and, in the same pass, estimates binding affinity — approaching atomistic accuracy while being orders of magnitude faster. MIT licensed, open weights, no access request.
Open model predicting 3D structures of proteins, DNA, RNA and ligands, plus binding affinity — near-atomistic accuracy, orders of magnitude faster. MIT-licensed open weights. Ships in a JupyterLab with a ready-to-run example.
What it is for
- Virtual screening of candidates by predicted affinity, in batch
- Protein-ligand complex structures for drug design
- Academic research without waiting on a shared cluster queue
- MIT license and open weights: reproducible, no gatekeeping
How to deploy
- Create your account and add balance (card or Pix, no subscription).
- In the console, pick the Boltz-2 template and a machine — the console hides the ones that do not meet the requirement.
- In about 15 minutes the setup finishes and the access address shows up in the panel, on port 8888.
Done? Just destroy the machine and billing stops with it. No contract, no minimum commitment.
FAQ
How is it different from AlphaFold 3?
AlphaFold 3's weights and training details are restricted; Boltz-2 is MIT and open, and it also predicts binding affinity, which the other does not deliver.
How much video memory do I need?
24 GB covers most targets. Very large complexes want a bigger card.
Does my research data leave the machine?
Only if you use the public multiple-sequence-alignment server, which is optional. Generate the alignment locally and nothing leaves the instance.