DeepHermes

by Nous Research

About the model:

Hybrid reasoning model based on Llama.

Model type:

Text

Model performance class:

Full

Link to the model:

https://huggingface.co/NousResearch/Hermes-4-70B

Base models:

Llama-3.1-70B

End model:

Hermes-4-70B

End model license:

Llama 3 Community License Agreement

About the organisation:

AI research initiative.

Link to the organisation:

https://nousresearch.com/

Model release date:

August 2025

Availability

Base Model Data

based on LLama3

End User Model Data

Mix of synthetic data. Data for previous model version made available.

https://huggingface.co/datasets/NousResearch/Hermes-3-Dataset

Base Model Weights

Inspecting the training weights requires signing Meta Llama 3.1's bespoke 'community license', not an OSI recognised open license

https://huggingface.co/meta-llama/Meta-Llama-3.1-70B

End User Model Weights

Model made available through HuggingFace.

https://huggingface.co/NousResearch/Hermes-4-70B

Training Code

Training code published on GitHub.

https://github.com/NousResearch/atropos

Documentation

Code Documentation

Code comprehensively documented.

https://huggingface.co/datasets/NousResearch/Hermes-3-Dataset

Hardware Architecture

Hardware setup comprehensively documented in paper.

https://arxiv.org/pdf/2508.18255

Preprint

Preprint published on arXiv.

https://arxiv.org/pdf/2508.18255

Paper

No peer-reviewed paper found.

Modelcard

Model card primarily contains usage information.

https://huggingface.co/NousResearch/Hermes-4-70B

Datasheet

No datasheet found.

Access

Package

No package found.

API and Meta Prompts

No API found.

Licenses

Meta custom license

https://huggingface.co/datasets/NousResearch/Hermes-3-Dataset

Is this information not up to date?

Contribute here ->

Supported by the Centre for Language Studies and the Dutch Research Council. Website design & development © 2024 by BSTN. This version of the index generated 10 March 2026, website content last updated 11 March 2026.