In recent years, artificial intelligence (AI) has become an integral part of technological progress. However, the dominance of large corporations in AI model development raises questions about the accessibility and control of these technologies. Nous Research — a startup founded in 2023 in New York, aims to change this order by offering decentralized, open, and human-centered AI models. The company is actively developing projects such as Psyche and Hermes 3 and has attracted significant investments to realize its ambitious goals.
Contents:
- Overview of Nous Research
- Psyche Project: Decentralized AI Training
- Hermes 3 Models: Open and Human-Centered Solutions
- Forge Reasoning API: Enhancing AI Reasoning Capabilities
- Funding and Future of Nous Research
- Conclusion
Overview of Nous Research
Nous Research is a research group focused on creating "personalized, unlimited" AI models. The company focuses on developing human-centered language models and simulators, aiming to align AI systems with real user experiences. Key areas of work include model architecture, data synthesis, fine-tuning, and reasoning, prioritizing open, human-compatible models that challenge the traditional closed approach.
Founded in 2023, the company is based in New York. Since its founding, Nous Research has attracted attention due to its unique approach to AI development, which includes the use of blockchain technologies for decentralized training and community participation.
Psyche Project: Decentralized AI Training
The Psyche project by Nous Research is an innovative platform for decentralized artificial intelligence training built on the Solana blockchain. The goal of the project is to democratize AI development, allowing participants worldwide to jointly train models using their computing resources.
Main features of Psyche:
- Decentralized training: Psyche uses DisTrO (Distributed Training Over-The-Internet) technology, which allows efficient model training by distributing computing tasks across various network nodes. This reduces bandwidth requirements and makes the training process more accessible.
- Integration with Solana: The Solana blockchain coordinates and motivates participants, offering a reward system for providing computing power. This fosters a sustainable and scalable ecosystem for AI training.
- Flexibility and resilience: The system demonstrates high resilience to failures, allowing new computing nodes to be dynamically added during training without loss of performance.
- Openness and accessibility: Psyche is focused on open-source code and accessibility for a wide audience, including researchers, developers, and AI enthusiasts.
Psyche represents a step toward a more open and inclusive future in AI, providing opportunities for collaborative learning and AI technology development on a global scale.
Hermes 3 Models: Open and Human-Centered Solutions
Hermes 3 is a series of language models developed by Nous Research, focused on high expressiveness, long reasoning, and individual customization. Hermes 3 models are available in the open and are used in various applications, including social networks, messengers, and gaming environments.
One feature of Hermes 3 is the ability to switch between intuitive and reasoning modes, allowing users to choose between quick answers and deeper, well-reasoned responses. This is achieved through technologies such as Chain of Code (CoC) and Mixture of Agents (MoA), which provide flexibility and accuracy in the model's outputs.
Forge Reasoning API: Enhancing AI Reasoning Capabilities
The Forge Reasoning API provides advanced reasoning features for language models, allowing them to effectively solve more complex tasks. One key component of the API is Monte Carlo Tree Search (MCTS), which helps build decision trees for planning tasks. Another important technology, Chain of Code (CoC), improves the model's capabilities in mathematics and programming through a series of sequential reasoning based on a code interpreter. Finally, Mixture of Agents (MoA) enables multiple models to work together, synthesizing new and more complete answers.
| Technology | Description |
|---|---|
| Monte Carlo Tree Search (MCTS) | Used for building decision trees, especially useful for planning. |
| Chain of Code (CoC) | A series of reasoning based on a code interpreter, improving capabilities in math and programming. |
| Mixture of Agents (MoA) | Allows multiple models to work together, synthesizing new and more accurate answers. |
Each of these technologies helps improve the accuracy and diversity of the solutions provided by the language model, making the Forge Reasoning API a powerful tool for complex computational tasks.
Funding and Future of Nous Research
In April 2025, Nous Research raised $50 million in a Series A funding round led by venture firm Paradigm. This round valued the company at $1 billion based on tokenized value. Earlier, in January 2024, the company raised $5.2 million in a Seed round from investors such as Distributed Global, North Island Ventures, and Delphi Digital.
With the raised funds, Nous Research plans to expand its computing resources, continue developing open AI models, and strengthen its position in the decentralized AI market. The company is actively developing projects such as Psyche and Hermes 3 and aims to create more accessible and ethical AI for everyone.
Conclusion
Nous Research is actively working on creating open and decentralized AI models, offering new opportunities for participants worldwide. The Psyche and Hermes 3 projects demonstrate a unique approach to training and fine-tuning models, making them more accessible and human-centered. With strong investor support and innovative technologies, the company is well-positioned to lead the market and open the way to an ethical and inclusive future of artificial intelligence.
