Evo 2 writes the book of life across all domains

Evo 2, a novel AI foundation model for biology developed by the Arc Institute, Stanford University, and NVIDIA, deciphers DNA grammar across all life domains with a record 1 million base pair context window. Trained on 9 trillion curated DNA base pairs, it achieves state-of-the-art zero-shot prediction of genetic variant effects, generates realistic genome-scale sequences, and features interpretable biological representations with cross-species transferability. Evo 2 also enables the design of DNA sequences with predefined chromatin accessibility patterns, validated in mouse and human cells. Fully open source with robust safety and equity design, it builds on prokaryote-focused Evo 1 and unlocks applications in human disease research and synthetic genome design. Though functional validation of generated sequences and computational efficiency need improvement, Evo 2 is a landmark advance toward a unified, cross-species model for reading, interpreting and writing genomic blueprints.