Hugging Face to Open-Source DeepSeek’s AI ‘Reasoning’ Model with Open-R1

Just a week after DeepSeek released its R1 reasoning model, researchers at Hugging Face have launched an ambitious effort to replicate and fully open-source the model under a new initiative called Open-R1. The project, led by Leandro von Werra and a team of engineers, aims to rebuild R1 from scratch and release all components, including training data, in the name of "open knowledge."

Why Hugging Face Is Rebuilding R1

While DeepSeek has released R1 with a permissive license, meaning it can be used without major restrictions, the model is not fully open-source—its training data, experimental details, and intermediate models remain undisclosed.

Elie Bakouch, an engineer on the Open-R1 project, explained the motivation behind their effort:

“The R1 model is impressive, but there’s no open dataset, experiment details, or intermediate models available, which makes replication and further research difficult. Fully open-sourcing R1’s complete architecture isn’t just about transparency—it’s about unlocking its potential.”

How Hugging Face Plans to Rebuild R1

The Open-R1 team will use Hugging Face’s Science Cluster, a powerful research infrastructure equipped with 768 Nvidia H100 GPUs, to generate datasets similar to those used by DeepSeek. The team has also opened the project to the AI and tech communities via Hugging Face and GitHub, encouraging contributions to the training pipeline.

The project has already attracted massive interest, gaining 10,000 GitHub stars in just three days. According to von Werra, crowdsourcing expertise will be key:

“A community effort is perfect for tackling [this], where you get as many eyes on the problem as possible.”

What Makes R1 Significant?

R1 is part of a new wave of reasoning models, which take longer to generate responses but are more reliable in fields like physics, science, and math. Unlike typical AI models, reasoning models “fact-check” themselves, reducing the likelihood of incorrect outputs.

DeepSeek’s rapid development of R1—released just weeks after OpenAI’s o1 model—has raised questions about U.S. leadership in AI, with some analysts worried that China’s AI advancements could outpace Western efforts.

However, the Open-R1 project is less about geopolitical competition and more about transparency in AI research. Bakouch emphasized the importance of controlling datasets and training processes for responsible AI deployment, particularly in sensitive fields like healthcare and law.

The Future of Open-Source AI

If Open-R1 succeeds, AI researchers could use its training pipeline as a foundation for future open-source reasoning models. Bakouch believes this approach benefits the entire AI community, including major labs and private AI companies:

“Rather than being a zero-sum game, open-source development immediately benefits everyone, including the frontier labs and model providers, as they can all use the same innovations.”

Despite concerns about open-source AI being misused, Bakouch argues that the benefits outweigh the risks, as more developers gaining access to reasoning models could accelerate AI innovation.

“When the R1 recipe has been replicated, anyone who can rent some GPUs can build their own variant of R1 with their own data, further diffusing the technology everywhere.”

What This Means

The Open-R1 project could push AI transparency forward, giving researchers unprecedented access to a cutting-edge reasoning model. If successful, this initiative may strengthen open-source AI development, providing an alternative to proprietary models while proving that major AI advancements don’t have to come exclusively from well-funded labs.

Additionally, this project comes at a time when Microsoft and OpenAI are investigating whether DeepSeek improperly used OpenAI’s API data to train its AI models. Since R1’s training data and methodology remain undisclosed, Open-R1’s attempt to "open the black box" could offer new insights into how DeepSeek trained its model. If Hugging Face researchers uncover similarities between R1 and OpenAI’s proprietary data, it could potentially back up Microsoft’s allegations.

On the other hand, if Open-R1 is built successfully without needing restricted or proprietary data, it could challenge claims that DeepSeek’s rapid development was due to unethical practices. Either way, this project could play a critical role in shedding light on how advanced AI reasoning models are being built—and whether corporate concerns over data misuse are justified.

Editor’s Note: This article was created by Alicia Shapiro, CMO of AiNews.com, with writing, image, and idea-generation support from ChatGPT, an AI assistant. However, the final perspective and editorial choices are solely Alicia Shapiro’s. Special thanks to ChatGPT for assistance with research and editorial support in crafting this article.