ESMC 6B (ESM Cambrian)
Release in the ESM (EvolutionaryScale and Biohub protein models) family · version ESMC-6B (esmc-6b-2024-12 on the Biohub Platform API)
Maintained by Biohub (Chan Zuckerberg Biohub, Inc.)134, EvolutionaryScale (original developer; now part of Biohub)79
The largest ESM C (ESM Cambrian) protein language model: 6 billion parameters, 80 transformer layers, and 2.37e23 training FLOPs. EvolutionaryScale introduced ESM C in December 2024; Biohub's May 2026 release of ESMC, ESMFold2, and ESM Atlas included the 6B weights, and ESMFold2 is trained on top of a frozen ESMC 6B.1736
- Model hub: Model card (Hugging Face) (external site: huggingface.co)
- Repository: GitHub repository (Biohub/esm) (external site: github.com)
- Paper: ESMC and ESMFold2 preprint (bioRxiv) (external site: biorxiv.org)
Availability and license
Overall availability
Downloadable from Hugging Face without gating. The card lists uses prohibited by Biohub's Acceptable Use Policy as out of scope.12
Availability is separate from permission: read the license before using or redistributing.
The card's license field lists "mit" and "other", with the license link pointing to the repository's third-party notice, which lists the licenses of code dependencies such as PyTorch, flash-attn, and xformers. The repository README states the models are available under the MIT license and asks users to follow Biohub's Acceptable Use Policy. EvolutionaryScale's December 2024 ESM C post, as archived on 2024-12-05, released the 300M and 600M weights and offered ESM C 6B through its Forge API and AWS SageMaker, while the current version of that post states that all three sizes were released as open weights under the MIT license.15387
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 | Downloadable from Hugging Face without gating.12 |
| Inference codeIs code for running the model published? | Public | The card documents loading the model with Hugging Face Transformers, and the esm package provides the production inference path.13 |
| Training codeIs the code used to train the model published? | Unknown | No ESM C training code was identified in the pages reviewed. |
| Training-data informationPublic = the training data itself can be obtained. Partial = composition or sources are documented without full access. | Partial | The card names the sources (UniRef, MGnify, and JGI sequences clustered at 70% identity into 83M, 372M, and 2B clusters) but does not provide the training set itself.1 |
| Training recipeAre the training configuration and procedure documented in enough detail to follow? | Partial | The card describes two stages: 1 million steps at 512-token context with 64% metagenomic data, then 500,000 steps at 2,048-token context with 37.5% metagenomic data. Further details are in the preprint, which was not reviewed.1 |
| Evaluation materialsPublic = evaluation code or prompts that let others re-run the evaluations are published. Partial = results only. | Unknown | The card refers to the preprint for performance metrics; the preprint was not reviewed. |
What it is useful for
Run and use notes
- The card documents inference with Hugging Face Transformers (AutoModelForMaskedLM) and recommends a GPU; using the flash attention backend requires running the model in bfloat16. The context window is 2,048 tokens.1
Organization context
Provenance and derivatives
Other releases in the ESM (EvolutionaryScale and Biohub protein models) family
- ESM3 open small (esm3-sm-open-v1)Model-disclosure tier (USASI rubric v0.1): Open-weight
ESM (EvolutionaryScale and Biohub protein models) family overview
U.S. eligibility
Eligible · basis: U.S.-governed project
ESM C was introduced by EvolutionaryScale, PBC (New York address in its former terms of use, effective January 15, 2025), and the 6B weights are published by Biohub, whose repository license names Chan Zuckerberg Biohub, Inc. The IRS lists Chan Zuckerberg Biohub Inc in Redwood City, California as a 501(c)(3) organization.7101411
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.