Robots Atlas>ROBOTS ATLAS
Evo 2
AI Models

Evo 2

2
A genomic foundation model (DNA language model) from Arc Institute, modeling DNA at single-nucleotide resolution with up to 1M base-pair context.
✓ Active✓ Public access⚖ Open sourceScientific AI
Context window
do 1 mln par zasad
tokens
Parameters
7B / 40B (także warianty 1B i 20B)
parameters
Release date
19 February 2025
🏢Arc InstituteProducer🤝NVIDIATechnology partner🔬Stanford UniversityResearch lab
Access:DownloadAPIHostedDeployment:💻 Local☁ Cloud

Overview

Evo 2 is a genomic foundation model (DNA language model) developed by Arc Institute in collaboration with partners including NVIDIA and Stanford University. It models DNA sequences at single-nucleotide resolution with context up to 1 million base pairs, using the StripedHyena 2 architecture.

The model was trained autoregressively on OpenGenome2, a dataset of 8.8 trillion tokens spanning all domains of life. This enables genetic variant effect prediction (e.g. zero-shot for BRCA1), gene-regulation analysis and generation of novel genomic sequences.

Evo 2 is released open source under the Apache 2.0 license. Weights are available on HuggingFace in 1B, 7B, 20B and 40B parameter variants; the model can also run via NVIDIA NIM or a hosted API. The work is described in Nature (2026).

Classification
Scientific AI
Applications
GenomicsComputational biologyBioinformatics
Access & deployment
DownloadAPIHosted
LocalCloud
Weights: Open source
Key parameters
📏 Context: do 1 mln par zasad
🧩 Parameters: 7B / 40B (także warianty 1B i 20B)
✓ Fine-tuning
📥 Input: text

Technical specification

Context window
do 1 mln par zasad
tokens
Parameters
7B / 40B (także warianty 1B i 20B)
parameters
License
Apache 2.0
Hardware requirements
The 40B/20B/1B models require FP8 via Transformer Engine and an NVIDIA Hopper GPU (40B needs multiple H100s). The 7B models run in bfloat16 on a single supported GPU.
Features:✓ Fine-tuning
Modalities
⬇ Input
text
⬆ Output
text

Capabilities and applications

Native model capabilities
DNA sequence understanding
Classifying and interpreting features of DNA sequences: regulatory regions, splice sites, organism group, motifs and transcription-factor binding sites.
Category: other
Variant effect prediction
Assessing how DNA variants, including single-nucleotide changes, affect genome function and regulation.
Category: other
Gene regulation prediction
Predicting regulatory properties of DNA sequences, such as gene expression, chromatin accessibility and RNA splicing.
Category: other
Next-base prediction
Predicting the next bases (nucleotides) of a DNA sequence.
Category: other
Long context
Support for large context windows — tens to hundreds of thousands (or millions) of input tokens. Enables analysis of entire codebases, long documents, and many parallel conversations without losing earlier information. GPT-5.1 supports 400,000 tokens.
Category: language
Synthetic data generation
Generating synthetic datasets that preserve the statistical properties of the original — used for model training, testing, and privacy protection.
Category: structured_generation
Application domains
GenomicsComputational biologyBioinformatics

Technical architecture

Core Architecture
S2StripedHyena 2
Model Form
DLDNA language model

Sources and related pages