Evo2
#13 in Wetenschop- & Bio-KIarc-institute · v2 · siet 19. Februar 2025 (Preprint/Erstveröffentlichung); Peer-reviewed Publikation in Nature im März 2026 · 12× · tolest 30. Juni 2026
Evo 2 is a genomic foundation model developed by Arc Institute in collaboration with NVIDIA, Stanford, UC Berkeley, and UCSF for DNA, RNA, and protein modeling. It processes biological sequences at single-nucleotide resolution with a context window of up to 1 million tokens and was trained on more than 9 trillion nucleotides from over 128,000 genomes across all domains of life. The model comes in 1B, 7B, 20B, and 40B parameter variants, is based on the StripedHyena 2 architecture, and is fully open-source (weights, training/inference code, OpenGenome2 dataset) under an Apache 2.0 license. It is distributed via NVIDIA BioNeMo/NIM, Hugging Face, and a browser-based interface called Evo Designer, and is primarily used for research in genetics, drug discovery, synthetic biology, and agricultur
Features
| Deployment Model | Local inference (Docker/Singularity), Cloud API (NVIDIA Hosted API/NIM), Self-hosting via GitHub code & Hugging Face weights |
| Use Case Scope | Genomic prediction & design: mutation effects, genome synthesis, drug discovery, agriculture, industrial biotechnology |
| Integrations | NVIDIA BioNeMo Framework, NVIDIA NIM Microservice, Hugging Face Hub, Goodfire interpretability visualizer |
| License | Apache License 2.0 (model & code fully open source) |
| Platform | GitHub, Hugging Face, NVIDIA BioNeMo/NIM, Evo Designer (web interface) |
| Price | Free/open source (model weights & code); NVIDIA API access via trial terms, no publicly listed pricing found |
| Release Date | Preprint/launch February 19, 2025; Nature publication March 2026 |