We're releasing Carbon-A: an open model to discover new genes in DNA. With Carbon-A, we annotated genomes from 22k+ species and produced 566 million gene candidates, about 16× the number of gene annotations in the RefSeq dataset. We're releasing the candidates as a database, alongside the model. Initial wet lab experiments support previously unannotated gene candidates even in well-studied organisms. We think 2027 will be the year for biology. As AI labs move from math and coding into biology, we want scientists to have capable open models and resources they can build on without depending on the priorities and policies of a handful of closed labs. Today, we can sequence genomes faster than we can identify the genes they contain. Finding genes often requires experimental data or comparisons with well-studied species, leaving much of the tree of life poorly understood. Carbon-A predicts where genes are directly from DNA, identifying regions that contain instructions for making proteins. Even familiar genomes still have gaps. Across cats, Syrian hamsters, chickens, and the plant Arabidopsis, independent RNA experiments in the wet lab supported 239 gene candidates missing from RefSeq, a major reference annotation database. Our ambition is to make finding genes across life as accessible as AlphaFold made predicting protein structures.