Nous Research, a New York-based AI team known for its "personalized and unrestricted" language models, has launched a new model inference API. This marks a significant step for the somewhat unconventional AI organization, making its lauded language models accessible to a wider range of developers and researchers through a programmatic interface.
Unrestricted Models
Nous Research has consistently challenged the perceived constraints of large AI companies like OpenAI and Anthropic. Their approach reflects a strong sense of "libertarianism." Now, they're packaging this freedom into an API for broader access.
The initial API release features two flagship products: Hermes3Llama70B, a heavyweight general-purpose model based on Meta's Llama 3.1 architecture, and DeepHermes-38B Preview, a reasoning model released last month that flexibly switches between standard responses and detailed Chain-of-Thought (CoT). This offers developers both a "luxury package" and a "budget-friendly" option.
However, accessing this "AI express delivery" isn't as simple as placing an order. Nous Research uses a waitlist system. To compensate waiting users, they provide a $5 credit for each new account.
This approach serves a dual purpose. Technically, it helps manage potential surges in demand, given Nous's likely more limited GPU resources compared to larger companies.
Strategically, the limited availability creates scarcity, piquing curiosity and generating buzz.
Interestingly, despite its unconventional approach, Nous Research's API design mirrors that of OpenAI, including completions and chat completions interfaces.
This ensures seamless integration for developers familiar with OpenAI's API, allowing easy incorporation of Nous's models into their applications. This demonstrates Nous's commitment to its principles while maintaining pragmatic commercial considerations. User-friendliness is paramount.
From "Free Download" to "Paid Deployment": A Business Evolution
Just four months ago, Nous Research launched Nous Chat, its first user interface chatbot. Before that, they focused on releasing open-source models for local deployment.
Previously, users had to download code and run models locally—a time-consuming, complex, and potentially costly process. The API allows developers to use high-performance models without infrastructure concerns. This marks a significant step toward a more sustainable business model.
This API launch reflects Nous Research's efforts to balance open-source principles with commercialization. They publicly release model weights while generating revenue through commercial deployment. It's a delicate balance—maintaining the spirit of "freedom" while ensuring viability.
This hybrid model attracts diverse users: individual developers and researchers can still download and run models for free, while businesses prioritizing reliability, convenience, and performance can opt for the paid API.
Nous Research plans to expand its inference services, potentially adding models like Hermes2Pro (skilled in function calling) and their Psyche project. For AI startups innovating with open-source models, Nous Research's API offers a new choice, disrupting the existing landscape, potentially intensifying competition in the AI inference field and driving further technological advancement.
