Evo 2 is a genomic foundation model from Arc Institute that handles prediction and design tasks across DNA, RNA and proteins. It models biological sequences at single-nucleotide resolution, with near-linear scaling of compute and memory relative to context length. The model has 40 billion parameters, a context window of 1 megabase and training on more than 9 trillion nucleotides from eukaryotic and prokaryotic genomes. Evo 2 was published in Nature in March 2026, following Evo 1 in Science in 2024.
The audience is computational biologists, genomics and protein engineering researchers, and machine learning teams working on biological sequences. Because the model is released through GitHub and Hugging Face, users should be comfortable running models in code or fine-tuning them on their own data. Arc Institute also offers Evo Designer and a mechanistic interpretability visualizer for people who want to explore the model's behavior without building everything from scratch. It is a research resource rather than a polished consumer application.
A researcher can score the likely effect of a genetic variant by comparing sequence likelihoods across a long genomic region. A synthetic biology group can generate candidate DNA sequences and then test them in the lab. Protein engineers can pair Evo 2 with Arc's MULTI-evolve framework to speed up design-test-build cycles. Teams building virtual cell models can use it alongside Arc tools such as State and Stack. Interpretability researchers can inspect which sequence features the model has learned, using the visualizer, and compare behavior between Evo 1 and Evo 2.
