The Carbon Annotation Database contains 566 million candidate protein-coding genes predicted by Carbon-A, an open 1.2-billion-parameter model that finds genes directly from DNA. It covers genomes from over 22,000 species, including animals, plants, fungi, and protists. Each prediction links to its source genome, genomic coordinates, coding DNA, predicted protein sequence, and confidence score, helping researchers explore poorly annotated genomes and prioritize candidates for further study.