Tier 2 Restricted open-weight
What the publisher claims: open-source inference code, model parameters available
The one-liner: Open code, restricted weights: the sharpest licence contrast in the batch.
The six questions
Parameters are downloadable from Google public storage subject to the terms, in three published variants.
Custom terms restrict use to non-commercial organizations, forbid redistributing the weights, and forbid training similar models on the output.
The inference pipeline runs locally in Docker on user-owned GPU hardware.
The Nature paper and Supplementary Information describe the method; training data, filters and training code are not released.
A skilled group could build a comparable system from the paper, but not reproduce the released weights.
google-deepmind/alphafold3 shows a single merged PR; development is entirely internal.
License
Code Apache 2.0; parameters under AlphaFold 3 Model Parameters Terms of Use plus the Model Parameters Prohibited Use Policy — license source ↗
- Non-commercial use only, including research on behalf of commercial organizations
- Weights may not be published or shared outside the requesting organization
- Output may not be used to train similar structure-prediction models
- Revocable licence; access can be terminated on breach
What is actually released
- inference code
- model parameters (af3.bin.zst, plus SynthID and Leaving-Atom variants)
- output terms of use
Why it matters
Same lab, adjacent year, radically different permission profile, which is exactly what the word open has to survive.
Sources
- https://github.com/google-deepmind/alphafold3
- https://raw.githubusercontent.com/google-deepmind/alphafold3/main/WEIGHTS_PROHIBITED_USE_POLICY.md
- https://raw.githubusercontent.com/google-deepmind/alphafold3/main/WEIGHTS_TERMS_OF_USE.md
Derived from the openness dataset behind But Is It Open? — snapshot 2026-11-07 ↗. Scores are qualitative readings of the cited evidence, not verdicts; tier measures openness, not intelligence.