Arc and NVIDIA release Evo 2, an open DNA foundation model trained across all domains of life

Arc Institute/Evo genomic foundation models

Arc Institute and NVIDIA released Evo 2, a DNA model trained on over 9.3 trillion nucleotides from more than 128,000 genomes across bacteria, archaea, and eukaryotes. It can identify disease-causing mutations with >90% accuracy on BRCA1 variants, design entire new genomes up to bacterial length, and recognize biological patterns across disparate species.

• Training: 9.3 trillion nucleotides, 128,000+ complete genomes, all three domains of life.
• Capabilities: identify disease mutations (>90% on BRCA1), design genomes, cross-species pattern recognition.
• Processing: handles genetic sequences up to 1 million nucleotides simultaneously.
• Foundational technology open-sourced for therapeutic and engineering applications.