The paper introduces Evo 2, a cutting-edge biological foundation model designed to interpret and engineer life's genomic complexity across all domains. Trained on a massive 9 trillion base pair dataset called OpenGenome2, the model utilizes a sophisticated StripedHyena 2 architecture to achieve a high-resolution 1 million token context window. This scale allows Evo 2 to predict the functional consequences of genetic mutations and identify complex structural features, such as exon-intron boundaries and protein elements, without specific fine-tuning. Beyond analysis, the model demonstrates generative capabilities, producing coherent genomic sequences and experimentally valid biological patterns. Ultimately, the developers have released Evo 2 as an open-source tool to accelerate global research in synthetic biology and clinical genomics.

References:

• Brixi G, Durrant M G, Ku J, et al. Genome modelling and design across all domains of life with Evo 2[J]. Nature, 2026: 1-13.