Chinese AI firm DeepSeek has unveiled an updated version of its flagship reasoning model R1, now available on the popular machine learning platform Hugging Face. The announcement was made early Wednesday via a post on the company’s official WeChat account.

Commercial-Ready AI with Open Licensing

The new R1 model, released under the highly permissive MIT license, is open for both academic and commercial use. DeepSeek describes this release as a “minor” upgrade to the original model launched earlier this year, a version that garnered significant attention for its high performance in AI reasoning tasks.

Despite the release, the Hugging Face repository currently offers limited documentation. Developers will only find configuration files and model weights, without any detailed description of the model’s architecture or capabilities.

Hefty Hardware Requirements

Boasting a staggering 685 billion parameters, the updated R1 model is not lightweight. These parameters, synonymous with model "weights," define the model’s behavior and learning capabilities. Such a large-scale model is unlikely to run on consumer-grade hardware without substantial resources, pointing to its potential use in enterprise or research-grade environments.

A Rising Star and a Point of Contention

DeepSeek made headlines earlier in the year with the original R1 release, which challenged industry leaders such as OpenAI in benchmark performance. However, the company’s rapid growth and advanced AI capabilities have not gone unnoticed by international regulators.

In the United States, concerns have been raised regarding the potential national security implications of DeepSeek's technologies. Critics argue that access to such powerful AI models may have strategic consequences, further intensifying the ongoing debate around open-source AI development and cross-border tech innovation.