GNoME's 2.2 million predicted crystals may include useful compounds

AI system

System: GNoME (Graph Networks for Materials Exploration)
Model: Not assessed
Release: Not assessed
Application developer: DeepMind
Role: Not assessed
Configuration: Not assessed
Inputs and outputs: Not assessed
Autonomy: Not assessed
Workflow: Graph neural network trained on existing crystal databases with iterative active learning. High-throughput DFT stability checks of candidates. Open release of the predicted structures.
Human oversight: Not assessed
Relation to AI: The graph-network model is the vendor's. Its predictions are documented in a vendor-authored Nature paper and blog. Independent reviewers (Cheetham & Seshadri 2024; Kurlin 2026 as reported by C&EN) report near-duplicates, unrealistic compounds and little evidence of useful novel materials. Benefit therefore rests on the vendor's account and is contested.
AI contribution: Not assessed

Studies: Not assessed
Study design: Not assessed
Measurement: Not assessed
Setting and sample: Not assessed
Measures: 2.2 million crystal structures (380,000 stable)
Comparison: stable crystals previously identified by computational studies before GNoME
Results: Reference: 48,000 computationally stable materials. Result: a total of 421,000 stable crystals.
Additional value from AI: Not assessed
Limitations: Not assessed
Denominators: Not assessed
Materials for replication: Not assessed

Participants: Developer: DeepMind. Operator: DeepMind (dataset released openly; stable structures contributed to the Materials Project).
Jurisdictions: Benefit: Mountain View, California, USA. Developer: United Kingdom. Operator: United Kingdom.
Ownership and control: Not assessed
Dependencies: Not assessed
Cross-border connections: Not assessed
Geopolitical implications: Not assessed

Conditions for replication: Predicted structures are genuinely novel rather than near-duplicates of known entries. Predicted compounds are synthesizable and have useful properties. Independent experimental confirmation of utility.
Replication attempts: 2 follow-up entries. Benefit did not recur: 2.
Later findings: 8 April 2024. Cheetham and Seshadri's April 2024 perspective in Chemistry of Materials examined the GNoME listings and found scant evidence of novel, credible and useful compounds, identifying every one of ten sampled 'stable structure' entries in the ICSD; C&EN reported further duplicate findings by Kurlin in October 2026.
Availability: Not assessed
Institutional decision: Not assessed
Grounds: Not assessed
Action and result: Not assessed
Success criterion: Not assessed

Assessed benefits: People. Organizations. Society.
Depth: Not assessed
Reasons: People: No individual beneficiary. A dataset release. Organizations: Use by the Materials Project and A-Lab is reported only by the vendor. Independent reviewers find scant utility. Society: No demonstrated downstream technology. Novelty of the predictions is disputed.
Observed or intended: Not assessed
Rubric: Not assessed
Assessor and date: Not assessed
Alternative assessments: Not assessed
Fill: Independently studied

Kind: Not assessed
Related entries: Not assessed
Field: Materials science. Materials researchers.
Purpose: Predict stable inorganic crystal structures at scale to accelerate materials discovery
Dates: Date basis unassessed: 2023 (precision unassessed). Updated: 22 October 2026.
Stage: Not assessed
Version: SHA-256: 4b33996fd343…