Model Book/Boltz-1

Boltz-1

A-

MIT

Structure PredictionComplex Predictionopen
Updated 3 weeks ago
NextIn take

Boltz-1 is the open-source answer to AlphaFold3 — not as accurate on complexes, but you can actually deploy it. For most production teams, that tradeoff is obvious.

Specifications

ArchitectureDiffusion-based structure prediction with MSA conditioning
Parameters~300M
Training DataPDB, UniRef, BFD — fully open training pipeline
LicenseMIT license. Fully open weights, code, and training pipeline.
Hardware1x A100 40GB for most predictions
Inference CostSelf-hosted only — ~$0.02/prediction on cloud GPU
API AvailableNo
Weights AvailableYes

Benchmark Performance

BenchmarkScore
CASP150.82 GDT-TS (avg)
DockQ0.68 DockQ

When to Use This

  • +Production drug discovery pipelines where you need IP freedom and on-prem deployment
  • +High-throughput structure prediction at scale

When NOT to Use This

  • When absolute peak accuracy on complexes matters more than accessibility
  • Ligand docking — use Boltz-2 or DiffDock instead

Production Readiness

production

Known Users

  • Multiple pharma companies (undisclosed)
  • Academic structural biology labs

Grade Rationale

A-

The best open-source structure prediction model available. Within striking distance of AF3 on accuracy while being fully deployable, modifiable, and free. The gap on complexes is real but narrowing.

Sources

Update History

2026-03-20Initial entry