Boltz-2
Open-source structure and binding-affinity prediction, MIT licensed
What's new
Boltz-2 jointly predicts complex structure and protein–ligand binding affinity, approaching FEP accuracy ~1000× faster.
Boltz is the most widely adopted fully open alternative to AlphaFold 3. Boltz-1 was the first open model to approach AF3 accuracy; Boltz-2 goes further by jointly modelling complex structure and protein–small-molecule binding affinity.
Highlights
- MIT-licensed model, weights and training pipeline — free for academic and commercial use.
- Boltz-2 is the first deep-learning model to approach physics-based free-energy perturbation (FEP) accuracy while running roughly 1000× faster.
- Simple YAML inputs describe the biomolecules and the properties you want to predict.
Getting started
1pip install boltz -U2boltz predict input.yaml --use_msa_server
Related tools
Free for academia
AI Models
AlphaFold 3
Predict the joint structure of proteins, nucleic acids and ligands
Web appModel weightsCommand line
Open source
AI Models
Protenix
High-accuracy open-source AlphaFold3 reproduction from ByteDance
Python libraryCommand lineModel weights
Open source
AI Models
BoltzGen
Toward universal binder design across modalities
Python libraryCommand lineModel weights