China’s DeepSeek chatbot defies US restrictions to stun AI world

Chinese startup turns the AI sector on its head with an open-source AI model capable of o1 reasoning but developed at a fraction of the price.

February 3, 2025
Matilda French

A new open-source reasoning model known as DeepSeek-R1 has turned the AI sector on its head. 

Unveiled by the Chinese AI startup DeepSeek, the R1 matches the performance capabilities of OpenAI’s market leading o1 model, yet it was developed at a fraction of the price: a mere $5.6 million. This amount is a drop in the ocean when compared to the billions invested by the American technology giant.

Such is the chatter surrounding the app that as of Monday 27 January 2025, DeepSeek-R1 had overtaken ChatGPT and other rivals to become the top-rated free application on Apple’s App Store in the United States, the United Kingdom, and China.

Researchers are especially excited about the R1 model's ability to tackle complex reasoning tasks, particularly in the fields of mathematics, coding, and natural language processing.

Said Dimitris Papailiopoulos, Principal Researcher at Microsoft’s AI Frontiers lab, after testing the model: “DeepSeek aimed for accurate answers rather than detailing every logical step, significantly reducing computing time while maintaining a high level of effectiveness.”

But capability and cost aren’t the only reasons the R1 is making waves. It’s also open source, meaning that any AI researcher or engineer around the world can work to enhance and refine the system for different applications. 

As Dr. Jim Fan, NVIDIA Senior Research Manager & Lead of GEAR (Generalist Embodied Agent Research) Group, summed up on X: “We are living in a timeline where a non-US company is keeping the original mission of OpenAI alive – truly open, frontier research that empowers all… DeepSeek-R1 not only open-sources a barrage of models but also spills all the training secrets.”

The R1’s development is even more remarkable given the current technological restrictions imposed on China by the United States, which include limitations on the quantity and sophistication of advanced chips available for sale to the nation.

These sanctions were intended to reduce access to the hardware necessary for progress in AI. Yet, the outcome has been the emergence of a revolutionary AI model that is substantially less expensive and demands far less computing power. In an instant, the widely held belief that the United States is at the forefront of the AI industry has been challenged.

The news comes just days after a consortium of American tech firms and foreign investors announced The Stargate Project, a $500 billion initiative to build AI infrastructure across the United States.

Enjoy the power of multiple LLMs via a single chat interface with Narus.

Narus logo