Carbon-A proposes 566 million gene candidates across 22,617 species

Noted

Lab claim

Hugging Face released a gene-finding model and a database of 566 million predicted protein-coding loci.

How to read this claim

Measurement — An Iso-Seq readout supports some predictions as transcription. It does not show protein.
Bill — A score is not a price. If you will call a model next, list input and output per million tokens on the vendor page.
Door — Status on this card: Shipped.

Keep the habit · field card with prices and worked checks

What changed

Annotation of a new genome was a slow specialist job. Carbon-A proposes 566 million protein-coding loci across 22,617 species. Those loci are candidates with a confidence score, not confirmed genes.

What it introduces

A large public set of predicted protein-coding loci, each with a confidence score. They are candidates, not confirmed genes.

Who benefits Genomics labs, groups annotating poorly studied species, and people training biological models. Engineering, product, and security teams do not need to change a plan this week.

Try this

No action for most readers. If you use the database, filter on the confidence score and do not cite a locus as a confirmed gene.

What we do not know

An Iso-Seq readout supports some predictions as transcription. It does not show protein.

Evidence

Primary Hugging Face. Benchmarks reported by Hugging Face; not independently verified. A later RNA readout supports some predictions and does not show protein.

Independent The F1 scores and the database counts are Hugging Face's. An Iso-Seq readout supports some predictions as transcription. It does not show protein.

Sources

Hugging Face Primary · October 8, 2026
Weights

Still unclear? · Start here · ranks, prices, unknown An Iso-Seq readout supports some predictions as transcription. It does not show protein.

The record

What happened

On October 8, 2026, Hugging Face released Carbon-Annotator (Carbon-A), a 1.2-billion-parameter model that predicts protein-coding regions from DNA, and the Carbon Annotation Database built by running it across GenBank. The database currently holds 48,167 assemblies from 22,617 taxa and 566 million predicted protein-coding loci. About half of the target set is annotated, with another batch planned in three weeks. Weights are at HuggingFaceBio/Carbon-A-1.2B.

Why it matters

Sequencing got cheap. Annotation did not. A single open model that proposes genes across many eukaryotic groups is a real scientific release, and the honest label on each row is still predicted. Treat the 566 million loci as candidates with a confidence score, not confirmed genes. The result that earns attention is a disagreement with the reference plus an independent RNA readout, with protein evidence still ahead.

The model uses a 98,304-base-pair window and predicts on both strands for mammals, other vertebrates, invertebrates, plants, fungi, and protists. Across 42 benchmark genomes, Hugging Face reports a macro-averaged nucleotide F1 of 0.944, ahead of the baselines they ran: AUGUSTUS, Helixer, Tiberius, ANNEVO, OrionGeno, SegmentNT, and NTv3. Hugging Face says the database covers about 11 times as many taxa and 9 times as much sequence as the RefSeq-derived training corpus, about 27 trillion base pairs. A confidence score on each gene reached an AUROC of 0.876 for separating exact coding-sequence matches from other predictions, in their benchmark. Those figures are Hugging Face’s.

The authors note that RefSeq labels are imperfect and change over time. They report that an animal-only checkpoint still annotated plant genomes it had not trained on, and that on Tetrahymena thermophila, which uses a nonstandard genetic code, nucleotide F1 stayed at 0.960. On the pig, the model scored better against a 2026 assembly than against the older 2017 assembly it had seen in training, which the authors read as evidence it learned coding regularities rather than memorizing one annotation.

Initial wet-lab work with ActiveSite and UCSD used PacBio Iso-Seq on cat, Syrian hamster, chicken, and Arabidopsis. Iso-Seq supported some coding regions Carbon-A predicted that are absent from RefSeq. Hugging Face is explicit that this is evidence of transcription, not yet of translation into protein. Ribosome profiling and mass spectrometry are the stated next tests. Hugging Face says EMBL-EBI expects to carry the annotations, and that additional wet-lab results exist and are not public yet.