Is NovaAtom-Lite state of the art?Pending benchmarks.Announced Nov 10, 2026; no independent results yet.
Geodesic Intelligence's all-atom structure prediction model. Given the sequences of a biomolecular assembly's protein, DNA and RNA chains, together with its small-molecule ligands, it predicts the position of every heavy atom, with confidence for each residue and each pair of residues. Announced November 10, 2026 and available through the Geodesic API (model novaatom) and NovaDDE. Open weights were not stated, and no public weight release was found.
The details so far
- Lab
- Geodesic Intelligence
- Announced
- Nov 10, 2026
- API
- Live · novaatom
- Open weights
- Closed · Not stated; no public Hugging Face or GitHub weight release found
- Type
- All-atom structure predictor: a recycled pairformer trunk reads sequence, evolutionary alignments and optional templates; a diffusion process over atoms generates coordinates conditioned on that representation (Geodesic model page and release article)
- Molecule classes
- Protein, DNA, RNA and ligands (small molecules, cofactors, metals and ions, by chemical component code or SMILES), modelled together in one prediction (Geodesic model page)
- Training data cutoff
- Structures released up to September 30, 2021, the same cutoff as AlphaFold 3 (Geodesic release article)
- Outputs
- All heavy-atom coordinates (five independent samples per target, ranked), plus pLDDT, PAE, PDE, pTM/ipTM and a distogram (Geodesic model page)
- Hosted access
- Geodesic API at platform.geodesiclab.com (model novaatom) and NovaDDE. Pricing: a free plan opens the model catalog with 100 credits per week (Geodesic API and pricing pages)
- License
- Not stated
- Related
- Separate from a larger NovaAtom model Geodesic says is still in training, and from NovaDDE (the workbench that can call this model)
Tags marked unsure are not stated on the lab's own pages.
What would let us call it
Verdicts use only independent boards we already track. Two consecutive snapshots are required before any crown.
- Publicly announced with a source
- Usable by the public (API or weights)
- Listed on an independent board we track
- Enough results for a rank in a category
- Holds across two snapshots (crowns only)
What the lab claims
Self-reported. Shown for context; not used in any verdict.
Lab claim: self-reported on FoldBench, 79.07% of 172 antibody–antigen interfaces predicted correctly (DockQ ≥ 0.23) by the top-ranked prediction from five seeds of five samples (Geodesic release article).
Lab claim: self-reported on FoldBench, 77.06% success on 558 protein–ligand interfaces (ligand RMSD < 2 Å and LDDT-PLI > 0.8) by the same top-ranked protocol (Geodesic release article).
Sources
LABGeodesic Intelligence: Introducing NovaAtom-Litegeodesiclab.com/news/introducing-novaatom-lite · Nov 10, 2026LABGeodesic Intelligence: NovaAtom-Lite model pagegeodesiclab.com/models/novaatom · Nov 10, 2026LABGeodesic API docsgeodesiclab.com/apiLABGeodesic Intelligence on X: Introducing NovaAtom-Litex.com/Geodesiclab/status/2120214931731775587 · Nov 10, 2026Tracking continues until an independent board lists it.