DeepSeek Opens Access To AI Code, Expanding Open-Source Efforts

China’s DeepSeek has made waves by releasing its AI models as “open source.”

The move raises questions about what this means for projects trying to replicate DeepSeek’s achievements. I Photo: Tim Reckmann Flickr

However, in the AI world, that term can mean different things. While DeepSeek previously allowed free use and modification of its models, it had yet to publish the underlying code—until now, David Meyer reported for Fortune’s Data Sheet.

Last week, DeepSeek announced that it would open-source five of its code and data repositories starting this week.

“These humble building blocks in our online service have been documented, deployed, and battle-tested in production,” DeepSeek stated.

“As part of the open-source community, we believe that every line shared becomes collective momentum that accelerates the journey. Daily unlocks are coming soon. No ivory towers—just pure garage energy and community-driven innovation.”

The move raises questions about what this means for projects trying to replicate DeepSeek’s achievements—such as its novel method of compressing large AI models for low-cost training.

“That depends on what exactly DeepSeek releases,” said Lewis Tunstall, a researcher at Hugging Face working on the Open-R1 project.

“We will be very excited if DeepSeek open-sources its reinforcement learning pipeline, as users are grappling with technical challenges like scaling training for very large models.”

Tunstall added: “It's unclear whether they will be open-sourcing code for inference, training, or something else entirely, but in either case, it’s a net win for the open-source community. In general, open-source development benefits from having multiple approaches to the same problem.”