New York artificial intelligence team Nous Research, known for its "personalized, unlimited" language model, recently launched a brand new model reasoning API. This move marks the AI organization, which has always been known for its "unruly", that it has begun to provide its proud language model to a wider range of developers and researchers through programmatic interfaces.
Nous Research has always challenged the "shackles" of large AI companies such as OpenAI and Anthropic, and their slogans are full of "liberalism". Now, they have decided to package this "freedom" into an API so that more people can experience it. The first batch of APIs are equipped with two celebrity products: the heavyweight general model Hermes3Llama70B based on the Meta Llama3.1 architecture, and the DeepHermes-38B Preview released last month, which can flexibly switch between standard reply and detailed "thinking chain".
However, trying this "AI Express" is not as simple as placing an order directly. Nous Research uses a first come first served waitlist system and provides a free $5 credit for each new account. On the one hand, this operation is to deal with the possible peak demand, and on the other hand, it is also a smart marketing method, creating a "sense of scarcity", making people more curious about what they are selling.
Interestingly, although Nous Research seems to be "unique", it chooses to align with OpenAI's API design pattern in API design, including completions and chat completions interfaces. This is a seamless connection for developers who are already familiar with the OpenAI interface, and can easily integrate Nous' models into their applications. This reflects that while Nous adheres to his own philosophy, he also shows pragmatic business considerations.
Just four months ago, Nous Research just launched their first user interface chatbot, Nous Chat. Prior to this, they have been focusing on releasing open source models for users to deploy locally. In the past, if users wanted to deploy these models, they needed to download the code themselves and run it locally. This was undoubtedly a time-consuming, tedious and possibly expensive "physical work". Now, through APIs, developers can directly use high-performance models without worrying about infrastructure. This marks an important step forward for Nous Research from a purely open source model to a more sustainable business model.
The launch of the API reflects Nous Research's efforts to pursue commercialization while adhering to the open source principle. While exposing model weights, they generate revenue through commercial deployment. This hybrid model is obviously intended to attract different user groups: individual developers and researchers can still download and run models for free, while those pursuing reliability, convenience and performance optimization can choose to use the API for a paid time.
Over time, Nous Research said their inference service will continue to expand and may include more featured models, such as Hermes2Pro, which is good at function calls, and their Psyche project. For AI startups that innovate based on open source models, Nous Research’s API undoubtedly provides a new option, breaking the existing landscape, potentially exacerbating competition in the field of AI reasoning and driving further technology.
Entrance: https://portal.nousresearch.com/login