Tier 4 Deeply open
What the publisher claims: open-source code, open weights, open database
The one-liner: Downloadable code, parameters and database, with share-alike databases and no training code.
The six questions
Inference code and model parameters are downloadable, and the AFDB is published under CC BY 4.0.
Apache 2.0 code and CC BY 4.0 parameters allow use, modification and redistribution with attribution, but the BFD and UniRef30 mirrors carry CC BY-SA 4.0 share-alike.
Runs locally in Docker on any modern NVIDIA GPU; ports such as ColabFold and OpenFold are widely used.
The Nature paper and technical note document architecture, training data and pipeline in detail, but no training code or checkpoints were released.
A skilled group can build a substantially equivalent predictor from the paper and open databases, but cannot retrain the released model exactly.
DeepMind's alphafold repo has zero merged lifetime PRs — the code is a one-way publication.
License
Code Apache 2.0; parameters CC BY 4.0; mirrored databases mixed (BFD and UniRef30 CC BY-SA 4.0, MGnify and PDB CC0); AlphaFold DB CC BY 4.0 — license source ↗
- Share-alike obligations on BFD and UniRef30 mirrors
- Attribution required for parameters and AFDB entries
What is actually released
- inference code
- model parameters
- AlphaFold Protein Structure Database
- mirrored genetic databases
- technical note v2.3.0
Why it matters
The reference case for open in science: nearly everything needed to use and audit the system is downloadable, yet the training pipeline stays closed.
Sources
- https://alphafold.ebi.ac.uk/
- https://github.com/google-deepmind/alphafold
- https://raw.githubusercontent.com/google-deepmind/alphafold/main/README.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.