AI Potluck
Back to Gap Map Model components / Scientific AI models

Evo 2

Arc Institute
open source / Overall score: 2.4

Evo 2 is Arc Institute's autoregressive DNA language-model family for genome modeling and sequence design. It works at single-nucleotide resolution with contexts up to one million base pairs, spans multiple model sizes, and provides local inference, sequence generation, embeddings, and documented training and fine-tuning paths.

Verified 2026-09-10 via the Hugging Face model card and Arc Institute repository.

Openness

5 high confidence
5.0
weights
open(Apache-2.0 ungated checkpoint)
data
open(OpenGenome2 raw FASTA and augmented JSONL released)
code
open(Savanna training framework and Evo 2 training guidance released)
license
Apache-2.0(OSI)

Evo 2 releases all three parts of the from-scratch stack: permissive weights, the OpenGenome2 training corpus, and the named open training framework.

Adoption

1 high confidence
1.0

1,350 downloads in the trailing 30 days for the declared 7B checkpoint.

Capability

3 high confidence
3.0

Two attributes. The multi-scale credit was withdrawn under ruling (e): 8K to one million bases is a longer window at the same single-nucleotide granularity, which is extent rather than a second scale, and it is the same reading that declines presto's 1-24 monthly timesteps and aifs's 13 pressure levels. Model-size variety is not an instrument.

Verified 2026-09-10