AlphaFold 3

A dated reading of what is claimed, reported, and independently verified in the current evidence.

Current reading

Readiness band and full integer triple

AI-assisted assembly · derived results

Readiness band

Ready

Primary summary from verified readiness

Evidence strength Critical

Decision

Agent-assisted assessment · independently reviewed · 2026-10-02

Use AlphaFold 3 for research structure hypotheses when critical predictions can be checked against experiments.

Why: Independent benchmarks support useful modeling, with uneven reliability for unfamiliar ligands, antibodies and nucleic acids. Ion-channel binding poses showed substantial failures. Clinical use is outside the model's declared validation. See [FoldBench](https://www.nature.com/articles/s41467-025-67127-3), the [channel study](https://www.frontiersin.org/journals/biophysics/articles/10.3389/frbis.2026.1937302/full) and the [official repository](https://github.com/google-deepmind/alphafold3).
Next: Pin the version and alignment pipeline, test representative held-out complexes, inspect chemistry and predictions across seeds, and experimentally confirm decision-critical poses. Choose the academic or commercial access route appropriate to the intended use.

Constraints

Accuracy can decline on chemically or structurally unfamiliar targets.
Antibody interfaces and complex nucleic-acid structures remain difficult.
Plausible protein scaffolds can accompany incorrect ligand-binding poses.
Confidence ranking can select an inaccurate prediction.
Hardware and alignment settings require validation for the chosen deployment.
Academic weights and outputs have non-commercial restrictions; commercial access requires a separate allowlisted subscription.
Clinical use is outside the declared validation.

Evidence summary

AI-assisted assembly

Counts and dates only. Raw signals, private excerpts, trust records, and internal corpus material are not published here.

Publication record

Revision 4