Nous Research
US company making the open Hermes AI models and the Hermes Agent
Profile
- Country
- United States
- Headquarters
- United States (no official head-office address published; office in New York)
- Founded
- 2023 (began in 2022 as an online volunteer group)
- Founders
- Jeffrey Quesnelle, Karan Malhotra, Teknium, Shivani Mitra
- Type
- Venture-backed corporation (Nous Research, Inc., Delaware)
Nous Research is an American AI research company. It began in 2022 as a group of volunteers who met on Discord, GitHub and Twitter and became a company in 2023. Its co-founders are Karan Malhotra, Jeffrey Quesnelle, a researcher who goes by the name ‘Teknium’, and Shivani Mitra. The legal entity is Nous Research, Inc., registered in the US state of Delaware, and it has an office in New York. Reports in April 2025 named Quesnelle as CEO, but a company post in October 2026 was signed by Dillon Rolnick as CEO.
Its best-known work is the Hermes series of language models, made by fine-tuning (training an already-trained model further to refine it) open models such as Meta's Llama. Hermes 3 405B, released in August 2024, is based on Llama 3.1 405B, and Hermes 4, released in August 2025, is a ‘hybrid reasoning’ model that can think at length before answering when needed; all of its weights (the trained model files) were published. It also researches training AI on scattered computers linked over the internet. In August 2024 it announced DisTrO, which it says cuts the data passed between GPUs (chips for AI calculations) by 1,000 to 10,000 times (company-reported), and in May 2025 it opened the Psyche network, which uses the Solana blockchain to coordinate training. Hermes 4.3, released in December 2025, was its first production model post-trained on that network.
In April 2025 it received a $50 million Series A investment led by Paradigm, a venture firm focused on crypto. The news site The Block reported that the round valued the company's yet-to-launch token at $1 billion. In February 2026 it released Hermes Agent, an AI agent (a program that carries out multi-step tasks for people) that creates new skills and builds up memory the more it is used, under the MIT licence. In October 2026 it announced $90 million in funding from investors including NVIDIA and Samsung, and TechCrunch reported it as a Series B at a $1.5 billion valuation. The company estimates that Hermes Agent has been cloned more than 24 million times (company estimate).
Sources for this profile (12) · as of Oct 31, 2026
- A Note on Our Fundraise · Nous Research
- Terms of Service · Nous Research
- Careers · Nous Research
- Exclusive: Crypto VC giant Paradigm makes $50 million bet on decentralized AI startup Nous Research at $1 billion token valuation · Fortune
- Paradigm leads $50 million Series A round for decentralized AI project Nous Research · The Block
- Nous Research confirms it hit $1.5B valuation, launches AI agents for business users · TechCrunch
- NousResearch/Hermes-3-Llama-3.1-405B model card · Hugging Face (Nous Research)
- Hermes 4 Technical Report · arXiv
- NousResearch/DisTrO: Distributed Training Over-The-Internet · GitHub (Nous Research)
- Democratizing AI: The Psyche Network Architecture · Nous Research
- Introducing Hermes 4.3: Local Intelligence Globally Trained · Nous Research
- NousResearch/hermes-agent · GitHub (Nous Research)
AIMade by AIThis profile was written by AI (Claude) from the 12 sources above (official pages, filings, news reports).How it's made
Records 10
Newest first. Open one to see it with its source.
The first controlled comparison showing distributed training is not just possible but equal or better
A US community team fine-tuning on top of a Chinese base model — a symbolic scene of shifting leadership
Distributed training moves beyond pretraining demos into the post-training path of real product models
An approach that benchmarks lower refusal rates as an alignment metric in their own right. The open community formalizes an alternative axis to safety alignment
Unlike the usual practice of opening only the weights, it released the post-training data too. A baseline for reproducibility in the open community
The largest pretraining run over the internet. Solved collaboration among untrusted participants with blockchain coordination
Claims at the level of a preliminary report turned into a reproducible paper and code
The starting point of the approach of training over internet bandwidth, without data center interconnects
The first case of the community actually fine-tuning the 405B model
The de facto standard recipe for context extension, unifying Position Interpolation and NTK-aware interpolation