# Boltz

> Open-source family of models for predicting biomolecular complex structures and binding affinities.

Boltz is a family of biomolecular interaction prediction models distributed as an open-source Python package and command-line tool. Boltz-1 and Boltz-2 are described in technical reports, and the code and weights are released under the MIT license. The latest listed release is v2.2.1, published 2025-09-08.

## What It Is

Boltz predicts the 3D structures of biomolecular complexes from YAML input files describing the molecules to model. Boltz-2, the latest model, jointly models complex structures and binding affinities. The authors say it approaches the accuracy of physics-based free-energy perturbation methods while running 1000x faster, which they present as making in silico screening practical for early-stage drug discovery.

## How Prediction Works

Users run `boltz predict` on a single YAML file or a directory of YAML files for batch processing. An option enables an MSA server for automatic multiple sequence alignment generation, with optional authentication. Affinity output has two values:

- `affinity_probability_binary` separates binders from decoys, for example in hit discovery.
- `affinity_pred_value` reports log10(IC50) and is meant for ligand optimization stages such as hit-to-lead and lead optimization.

## Setup and Hardware

Install from PyPI with the CUDA extra, or from GitHub source. A CPU-only install is supported but significantly slower. On recent NVIDIA GPUs, Boltz uses cuEquivariance kernels. A community fork runs it on Tenstorrent hardware.

## Current Status

The README says updated Boltz-2 evaluation and training code is coming soon. Boltz-1 training instructions are available now.

## Features
- Biomolecular complex structure prediction
- Binding affinity prediction
- Binder vs decoy probability output
- Batch inference from YAML files
- Automatic MSA generation via MSA server
- MSA server authentication
- CUDA acceleration with cuEquivariance kernels
- CPU-only mode
- MIT-licensed code and weights

## Integrations
ColabFold MSA server, NVIDIA cuEquivariance, Tenstorrent (community fork)

## Platforms
API, DEVELOPER_SDK, CLI

## Pricing
Open Source

## Version
v2.2.1

## Links
- Website: https://github.com/jwohlwend/boltz
- Documentation: https://github.com/jwohlwend/boltz/blob/main/docs/prediction.md
- Repository: https://github.com/jwohlwend/boltz
- EveryDev.ai: https://www.everydev.ai/tools/boltz