Hugging Face Bio releases the Carbon Annotation Database and Carbon-A, a 1.2B-parameter gene-annotation model (Apache-2.0)

Hugging Face Bio released the Carbon Annotation Database and the model behind it, Carbon-A (1.2B parameters, Apache-2.0), which predicts protein-coding sequence at single-base resolution in eukaryotic genomes. The team used it to propose 566 million new gene candidates across 22,617 species; a number were validated in cats, chicken and arabidopsis with partners at Active Site and UCSD. Weights, annotation pipeline, database, training data and technical report are in the HuggingFaceBio collection. Blog post of 8 October 2026; accuracy and discovery counts are the authors' own and eukaryote-only.