Teams publishing in Nature unveiled Evo 2, a DNA foundation model trained on genomic data from over 100,000 species that the authors say can both interpret and generate long-form genomic sequences across life. The papers describe Evo 2’s ability to identify disease-causing human mutations and to propose full synthetic genomes at bacterial scale, positioning the model as a platform for genome design and comparative genomics. A foundation model in genomics applies large-scale machine learning to capture statistical patterns in sequence data that can be repurposed for prediction or design tasks. Authors caution that experimental validation remains necessary before designed sequences are used in the lab or clinic, but the capability signals accelerated computational design workflows for synthetic biology and biotech R&D.