Can you use AlphaFold 3 commercially?
Maintained by the Talindrew team · Last updated
Not with the public weights. The AlphaFold 3 code is Apache 2.0, but Google DeepMind's terms for the model parameters allow use only by or on behalf of non-commercial organisations such as universities, non-profits and government bodies, and they prohibit use in connection with commercial activities, including research done on behalf of a company. AlphaFold Server is also non-commercial. For commercial work, Google Cloud offers AlphaFold 3 to allow-listed customers under a commercial subscription, and several open AlphaFold 3-class models are licensed for commercial use: Boltz-2 (MIT), Chai-1 (Apache 2.0), OpenFold3 (Apache 2.0), Protenix v1 (Apache 2.0) and ESMFold2 (MIT).
What exactly do the AlphaFold 3 terms allow?
The code in google-deepmind/alphafold3 is Apache 2.0. The model parameters are a separate download under their own terms of use, which is what governs whether you may run predictions.
- Allowed: use by, or on behalf of, non-commercial organisations (universities, non-profit research, education, journalism, government).
- Not allowed: use in connection with any commercial activity, including research on behalf of commercial organisations.
- Not allowed: training other structure prediction models on AlphaFold 3 output, clinical use, or sharing the parameters outside your organisation.
- Allowed: publishing and sharing predictions you made under the terms.
What about AlphaFold Server?
AlphaFold Server, Google DeepMind's free web front end, is also for non-commercial use. As of January 2026 it allows 30 jobs a day of up to 5,000 tokens each, supports a more limited set of ligands and modifications than the full model, and has no API, so it cannot be built into a pipeline.
Is there any legal commercial route to AlphaFold 3?
Yes: Google Cloud lists AlphaFold 3 as generally available to an allow-list of customers, deployed to the customer's own endpoint under a commercial subscription arranged with a Google Cloud account team. Whether that subscription lets a vendor resell predictions to its own users is not stated publicly, so ask in writing before building a product on it.
Which AlphaFold 3-class models can you use commercially?
Several open reproductions and successors predict proteins with ligands, nucleic acids and modifications under commercial licences. Check each licence for the exact version you deploy: Protenix v2's weights, for example, are proprietary even though v1's are Apache 2.0.
AlphaFold 3 and the commercially usable alternatives (checked 2026-10-02)
| Model | Code licence | Weights licence | Commercial use |
|---|---|---|---|
| AlphaFold 3 (public weights) | Apache 2.0 | Google DeepMind terms, non-commercial | No |
| AlphaFold 3 on Google Cloud | n/a | Commercial subscription, allow-list | Yes, per contract |
| Boltz-2 | MIT | MIT | Yes |
| Chai-1 | Apache 2.0 | Apache 2.0 | Yes |
| OpenFold3 | Apache 2.0 | Apache 2.0 | Yes |
| Protenix v1 | Apache 2.0 | Apache 2.0 | Yes (v2 weights are proprietary) |
| ESMFold2 | MIT | MIT | Yes |
| HelixFold3 | Custom | Non-commercial | No (paid server only) |
Frequently asked questions
Is the AlphaFold 3 code open source?
The code is Apache 2.0, but the code alone predicts nothing: the trained model parameters are needed, and they are under separate non-commercial terms.
Can a startup use AlphaFold Server for drug discovery?
AlphaFold Server is for non-commercial use only, so work for a company falls outside it. Use a commercially licensed model such as Boltz-2 or ESMFold2, or ask Google Cloud about a commercial AlphaFold 3 subscription.
Which alternative is closest to AlphaFold 3 for protein-ligand complexes?
Boltz-2 co-folds a protein with a ligand and also predicts binding affinity, and OpenFold3 and Chai-1 reproduce AlphaFold 3's architecture. Benchmarks move every few months, so test on complexes like yours.
References
- Google DeepMind. AlphaFold 3 model parameters terms of use. https://github.com/google-deepmind/alphafold3/blob/main/WEIGHTS_TERMS_OF_USE.md
- Google DeepMind. AlphaFold 3 repository and README (code licence). https://github.com/google-deepmind/alphafold3
- Google Cloud. AlphaFold 3 in Model Garden: availability and commercial access. https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/open-models/alphafold-3
- Abramson J, Adler J, Dunger J, et al. Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature 630: 493–500 (2024). https://doi.org/10.1038/s41586-024-07487-w
- Passaro S, Corso G, Wohlwend J, et al. Boltz-2: Towards accurate and efficient binding affinity prediction. bioRxiv 2025.06.14.659707 (2025). https://doi.org/10.1101/2025.06.14.659707
- Chai Discovery. chai-lab repository (Apache 2.0). https://github.com/chaidiscovery/chai-lab
- ByteDance. Protenix repository and weight licences. https://github.com/bytedance/Protenix
- Biohub. ESMFold2 model card (MIT). https://huggingface.co/biohub/ESMFold2
Related
Design your Can you use AlphaFold 3 commercially? construct in Talindrew.
Free to use. No download required. Works in any modern browser.