AI Potluck
Back to Gap Map Model components / Scientific AI models

AlphaFold

Google
restricted / Overall score: 2.3

AlphaFold is Google DeepMind's biomolecular structure prediction line. AlphaFold 3, the current tier, predicts the joint structure of complexes of proteins, nucleic acids, ligands, ions and modified residues from sequence together with genetic-search and template inputs. The inference source is Apache-2.0 in both generations, but the model parameters are licensed separately, and separately in each generation.

The slug covers the product line at the tier DeepMind ships, which is AlphaFold 3 in google-deepmind/alphafold3; the declared repository is the AlphaFold 2 codebase the line began with, and it is still the more widely starred of the two. Verified 2026-09-10 via both GitHub repositories, the AlphaFold 3 weights terms of use, and the AlphaFold 2 README.

Openness

2 high confidence
2.0
weights
open(AlphaFold 3 parameters at a published URL but only if received directly from Google
data
closed(training corpus not released in either generation)
code
partial(Apache-2.0 inference pipeline in both repositories
license
AlphaFold-3-Model-Parameters-Terms-of-Use(non-commercial use only

The line ships two generations under two parameter licenses, and the multi-SKU rule resolves it on the distributed set. AlphaFold 2's parameters are CC-BY-4.0, which alone would not cap anything; AlphaFold 3's are also distributed, and their terms forbid commercial use and require that you receive them directly from Google. The most restrictive distributed license governs, so the current tier decides the score and the line sits at restricted. Scoring the line at AlphaFold 2 would be scoring a superseded product.

Adoption

3 medium confidence
3.0

No Hugging Face artifact and no published download count for either parameter set, so stars are the last-resort signal. The declared repository carries 14,844; the AlphaFold 3 repository that now holds the shipping tier carries 8,538, and neither number is a usage measure.

Capability

2 high confidence
2.0

One readable attribute. The breadth people associate with AlphaFold 3 is breadth of entity types in a single prediction task, and the category's note is explicit that a long task list is not an adaptation pathway - the repository ships inference only, and the model predicts a fixed target rather than designing anything.

Verified 2026-09-10