Chinese Developers Improved DeepSeek With a New Version of  R1 Model

China’s DeepSeek has released an updated version of its R1 language model and released it on the Hugging Face platform under an open MIT license. The changes to the model are minimal, but it can now be used for free in commercial projects.

The Hugging Face repository does not yet have a detailed description of the model. Only configuration files and “weights” – numerical parameters that determine its behavior and capabilities. The updated R1 contains 685 billion parameters, which makes it extremely resource-intensive and, as TechCrunch notes, it is unlikely to be possible to run such a model on regular user computers without additional optimization.

DeepSeek first attracted attention earlier this year when it unveiled R1, a competitive and reasoning alternative to OpenAI. That success raised concerns among some U.S. regulators, who saw the Chinese startup as a threat to U.S. national security.