AlphaFold3
AGoogle DeepMind
AlphaFold3 sets the new ceiling on structure + complex prediction, but the restricted license and API-only access make it a non-starter for any team that needs to ship into a regulated workflow. Pair it with Boltz-1 for production work, use AF3 for benchmarking and exploration only.
Specifications
| Architecture | Diffusion-based, Evoformer + diffusion module |
| Parameters | ~600M |
| Training Data | PDB, UniRef, BFD, Mgnify, plus proprietary complex data |
| License | Free for academic use via API. Commercial use requires licensing from Google DeepMind. No weight distribution. |
| Hardware | API-only — no local deployment possible |
| Inference Cost | Free via AlphaFold Server (rate-limited), enterprise pricing undisclosed |
| API Available | Yes |
| Weights Available | No |
Benchmark Performance
| Benchmark | Score |
|---|---|
| CASP15 | 0.85 GDT-TS (avg) |
| PoseBusters | 77% valid |
| DockQ | 0.73 DockQ |
When to Use This
- +Benchmarking and validation against your own models
- +Exploratory structure prediction for novel targets
- +Complex prediction where accuracy matters more than throughput
- +Academic research without commercial constraints
When NOT to Use This
- −Production drug discovery pipelines requiring IP freedom
- −High-throughput screening where API rate limits are a bottleneck
- −Regulated workflows requiring on-premises deployment
- −Any application where you need to fine-tune or modify the model
Production Readiness
Known Users
- Academic labs worldwide
- DeepMind/Isomorphic Labs internal
- Pharma companies via enterprise agreements
Grade Rationale
A, not A+. Best-in-class accuracy and the broadest task coverage of any structure model, but the restricted license and lack of weights drop it from the top tier. A+ is reserved for models that are both technically dominant and accessible to the field that needs them.
Roadmap
AF3 Server expanding to support larger complexes. DeepMind has hinted at broader API access but no timeline for weight release. Isomorphic Labs is building proprietary extensions for internal drug discovery.