Nous Research, the lab behind the open Hermes models

A lab for open AI

Nous Research is a company (Nous Research, Inc.) with a strong presence in online open-AI communities on Discord, X and YouTube. Its official pages do not give a founding year or headquarters city. The organization calls itself “a pioneer in open AI training and research,” with a mission to “create and proliferate open access to intelligence.”

The Hermes models

The Hermes family is its best-known work. These are open-weight models produced by post-training base models released by other labs, notably Meta’s Llama 3.1. The Hermes 4 generation, released in August 2025, offers a hybrid reasoning mode: the model can produce a long chain of thought before answering, or answer directly. Nous Research highlights its models’ “steerability,” meaning how well they follow the user’s instructions, with fewer refusals, as well as tool calling and valid JSON output.

Other models round out the lineup, such as NousCoder-14B for coding and Nomos 1. Licenses depend on the base model: Hermes 4 70B, for example, follows the Llama 3 license.

Training differently: DisTrO and Psyche

What most sets Nous Research apart is its work on decentralized training. The DisTrO and DeMo reports describe methods for training a model across machines linked by an ordinary internet connection, drastically reducing data exchange. The Psyche Network, launched in May 2025, puts these ideas into practice: training runs coordinated by smart contracts on the Solana blockchain, with no human intermediary. Nous Research also publishes Atropos, a framework of environments for reinforcement learning.

Hermes Agent and Nous Portal

In February 2026, the organization launched Hermes Agent, an open-source AI agent under the MIT license that runs in the terminal or as a desktop app. Nous Portal acts as a single account to power it, giving access to Hermes models and many models from other labs. For a first look, the nousresearch.com site lists all releases. To compare Hermes with other open models, see our page on open-source models.