DeepSeek releases updated R1 model on Hugging Face under open license

DeepSeek AI

Chinese AI startup DeepSeek has quietly rolled out an updated version of its R1 reasoning model on Hugging Face, the popular developer hub for open-source AI. The announcement came via a WeChat post recently, and while DeepSeek described the release as a “minor” upgrade, it carries some weight — literally and figuratively.

According to a report by Tech Crunch, the updated R1 boasts 685 billion parameters, making it one of the largest publicly released models to date. It’s distributed under the permissive MIT license, which allows for commercial use — a rare move in an industry increasingly leaning toward closed ecosystems, as per the report.

Despite its open license, the model isn’t exactly plug-and-play for hobbyists. With its massive parameter count, R1 likely requires specialised hardware and won’t run on consumer-grade machines without significant modification.

Notably, DeepSeek hasn’t shared much documentation with the release. The Hugging Face repository contains only the configuration files and model weights — with no detailed explanation of its capabilities, benchmarks, or use cases.

Still, R1’s arrival on Hugging Face is likely to attract interest. When the original version dropped earlier this year, DeepSeek earned comparisons to top-tier players like OpenAI. It also drew attention from U.S. regulators concerned about the national security implications of exporting such advanced AI technology.

By open-sourcing a powerful model under a commercial-friendly license, DeepSeek is positioning itself as both a challenger to Western incumbents and a contributor to the open AI community — a balancing act that will be closely watched as global tensions around AI governance continue to rise.