# ESM3 open (esm3-sm-open-v1)

ESM3 open (esm3-sm-open-v1). Code: MIT; weights: MIT. Last verified 2026-11-14. Query directly over MCP or API.

## Code license

Permissive · MIT
Commercial use: Yes · derivatives: Yes · redistribution: Yes

> **License (MIT)**
https://raw.githubusercontent.com/evolutionaryscale/esm/main/LICENSE.md (2026-11-13)

> Copyright 2026 Chan Zuckerberg Biohub, Inc.
https://raw.githubusercontent.com/Biohub/esm/main/LICENSE.md (2026-11-13)

## Weights license

Permissive · MIT
Commercial use: Yes · derivatives: Yes · redistribution: Yes

> This repository is under a MIT [license](https://github.com/evolutionaryscale/esm/blob/main/LICENSE.md).
https://huggingface.co/EvolutionaryScale/esm3-sm-open-v1/raw/main/README.md (2026-11-13)

> This repository is under a MIT license
https://huggingface.co/biohub/esm3-sm-open-v1 (2026-11-13)

## Commercial use of ESM3 open (esm3-sm-open-v1)

- Code: Yes
- Weights: Yes
The code license and the weights license stand apart; a verdict on one does not carry over to the other.
Verdicts record what sources state and are not legal advice.

## Limits on outputs

- The function decoder was filtered for potentially harmful keywords.
  > The function decoder has been filtered for potentially harmful keywords.
  https://huggingface.co/biohub/esm3-sm-open-v1 (2026-11-13)
- The README states the models are available under the MIT licence and asks users to follow Biohub's Acceptable Use Policy.
  > Please follow our [Acceptable Use Policy](https://biohub.org/acceptable-use-policy/) when using the model.
  https://github.com/Biohub/esm/blob/main/README.md (2026-11-13)

## Training data

- The model card says the model is trained on 2.78 billion natural proteins, which with synthetic augmentation totaled 3.15 billion protein sequences, 236 million protein structures, 539 million function-annotated proteins and 771 billion tokens.
  > `esm3-sm-open-v1` is trained on 2.78 billion natural proteins. With synthetic data augmentation, this led to 3.15 billion protein sequences, 236 million protein structures, and 539 million proteins with function annotations, totaling 771 billion tokens.
  https://huggingface.co/EvolutionaryScale/esm3-sm-open-v1/raw/main/README.md (2026-11-13)
- Trained on 2.78 billion natural proteins, expanded by synthetic augmentation to 3.15 billion sequences, 236 million structures and 539 million function-annotated proteins (771 billion tokens).
  > esm3-sm-open-v1 is trained on 2.78 billion natural proteins.
  https://huggingface.co/biohub/esm3-sm-open-v1 (2026-11-13)
- Virus-related data and some proteins from organisms on the USDA Select Agents and Toxins list were removed from training data.
  > data related to viruses has been removed from the training dataset
  https://huggingface.co/biohub/esm3-sm-open-v1 (2026-11-13)

## How to run

- Weights download: https://huggingface.co/EvolutionaryScale/esm3-sm-open-v1
- Third-party hosting: https://github.com/Biohub/esm/blob/1b3e08025b/LICENSE.md

## Versions

- ESM-3: code —, weights —
- unspecified: code —, weights —
- 304: code —, weights —

## No sourced statement yet

Overlap with benchmarks, Known defects and misuse

https://mdexa.com/models/esm3-open/

## Continue in this database

This record is one of 109 processed and verified records (datasets, data resources and open models). To find others, call `https://mcp.mdexa.com/v1/datasets?q=<words>` or connect the MCP server at `https://mcp.mdexa.com/mcp`; no key is needed for 20 calls per minute. Rows are under 1 KB; an empty result means "not listed". For a list of items, one batch check (free account) returns a report. Guide: /agents.md. Similar records: ESM C (https://mdexa.com/models/esm-c/); AlphaFold 3 (https://mdexa.com/models/alphafold3/); Boltz (Boltz-1 and Boltz-2) (https://mdexa.com/models/boltz/).