DeepSeek, a Chinese startup, has unveiled an updated version of its R1 reasoning AI model on the developer platform Hugging Face, following an announcement made on WeChat earlier Wednesday.

According to DeepSeek’s WeChat announcement, the updated R1, which operates under a permissive MIT license allowing for commercial use, is considered a “minor” upgrade. However, the Hugging Face repository only provides configuration files and weights, the internal components that dictate a model’s behavior, without including a description of the model itself.

The updated R1 model has a substantial size of 685 billion parameters, which is equivalent to its weight. In its current form, it is unlikely to be compatible with standard consumer-grade hardware without modifications.

DeepSeek gained significant attention earlier this year with the release of R1, which demonstrated impressive capabilities comparable to those of OpenAI models. Nevertheless, the startup has also attracted scrutiny and raised concerns among some regulators, who argue that DeepSeek’s technology poses a potential national security risk.