AI provider

Nous Research

Developers of the Hermes model families and open training tools. Creators of Hermes Agent, software that works with models from multiple providers.

Company & leadership

Chief executive
Dillon Rolnick
Chief technology officer
Jeffrey Quesnelle
Legal entity
Nous Research, Inc., a Delaware corporation

Founders

  • Jeffrey Quesnelle
  • Karan Malhotra
  • Teknium
  • Shivani Mitra

An open research company

Nous formed in 2023. Its early work focused on post-training other developers' base models. It published Hermes models, synthetic datasets and YaRN research for longer context windows.

Founders and leadership

Shivani Mitra describes her role as founding CEO and president. She says she built the organization and raised $70 million from 2023 to 2025. Jeffrey Quesnelle identifies himself as co-founder and CTO. Teknium led model post-training before becoming the main architect and maintainer of Hermes Agent.

Funding and the business product

In October 2026, CEO Dillon Rolnick announced a $90 million raise. Named investors included NVIDIA, M12, Samsung, Robot Ventures, Union Square Ventures, Y Combinator and Menlo Ventures. The announcement set out plans for Hermes for Businesses.

Training across the internet

Hermes 4.3 was trained through the Psyche network. Nous also releases the Atropos reinforcement-learning framework and models that document specific training experiments. These research artifacts have different readiness levels from general chat releases.

What the model list covers

Practical guidance. The list covers documented Nous model releases and Teknium's OpenHermes family. It includes older weights that remain downloadable. Copies of unchanged Meta or other third-party models, tokenizer packages, dataset uploads and undocumented training checkpoints are not listed as new Nous models.

Model terms differ across releases

Practical guidance. The software license for Hermes Agent does not define the terms for every Nous model. Some older repository labels describe a fine-tune while the underlying model has separate terms. Check the release files and the base model terms for the package you select.