Evo 2 20B
Release in the Evo 2 family · version evo2_20b
Maintained by Arc Institute14
A 20-billion-parameter Evo 2 checkpoint with 1 million base-pair context, released in February 2026. Arc created it from Evo 2 40B without additional training by removing layers that logit-lens analysis showed could be dropped with little effect on loss, while keeping the unembedding layer.14
- Model hub: Model card (Hugging Face) (external site: huggingface.co)
- Release notes: Release v0.5.0 notes (external site: github.com)
- Repository: GitHub repository (external site: github.com)
Availability and license
Overall availability
Downloadable from Hugging Face without gating.12
Availability is separate from permission: read the license before using or redistributing.
The evo2 LICENSE file also carries the notices for bundled third-party code, including a BSD-style NVIDIA license and the MIT license for Fairseq code.6
Model-disclosure tier
The model parameters for this release can be downloaded by the public. License terms may still restrict use, redistribution, or commercial use.
Public materials checklist
| Item | Status | Notes and evidence |
|---|---|---|
| WeightsCan the general public download the model parameters for this release? | Public | Weights are on Hugging Face without gating; a Savanna-format checkpoint (savanna_evo2_20b) is also published.123 |
| Inference codeIs code for running the model published? | Public | The evo2 package on GitHub supports evo2_20b under Apache 2.0.56 |
| Training codeIs the code used to train the model published? | Partial | No further training was done; the weights come from Evo 2 40B, which the README says was trained with the open-source Savanna framework. Code for the layer-removal step was not identified.45 |
| Training-data informationPublic = the training data itself can be obtained. Partial = composition or sources are documented without full access. | Public | The card lists OpenGenome2, the Evo 2 pretraining dataset, which is downloadable from Hugging Face.17 |
| Training recipeAre the training configuration and procedure documented in enough detail to follow? | Partial | The release notes describe the layer-removal method at a high level; the exact layers removed were not found in the pages reviewed.4 |
| Evaluation materialsPublic = evaluation code or prompts that let others re-run the evaluations are published. Partial = results only. | Partial | The release notes compare the 20B and 40B checkpoints on variant effect prediction and gene completion; evaluation scripts for those comparisons were not identified.4 |
What it is useful for
Run and use notes
Organization context
Provenance and derivatives
Derived by Arc Institute from its own Evo 2 40B checkpoint through layer removal, without additional training.4
- Derived from: Evo 2 40B (evo2_40b) — Source checkpoint.
Other releases in the Evo 2 family
- Evo 2 40BModel-disclosure tier (USASI rubric v0.1): Open-stack
U.S. eligibility
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.