Chinese startup DeepSeek has announced that it will open source its models next week, reinforcing its dedication to open AI technology. The company shared on social media platform X that it will release five repositories, calling it “small but sincere progress” made with “full transparency.” According to DeepSeek, these components have been documented, deployed, and tested in a production environment.
Last month, DeepSeek introduced an open-source model, DeepSeek R1, capable of reasoning and performing at a level comparable to U.S.-based AI firms, despite its lower development costs. This has led investors to scrutinize AI companies’ cost reports more closely.
A Different Approach in China
Unlike most Chinese AI firms, which, like their U.S. counterparts, keep their models closed-source, DeepSeek is taking a different approach. Reuters notes that this strategy makes the company stand out in China’s AI landscape.
DeepSeek founder Liang Wenfeng stated in July last year that commercialization is not the company’s main focus. He sees open-source AI as a form of “soft power,” emphasizing the value of innovation being adopted by others. “When others follow your innovation, it gives you a deep sense of accomplishment,” he said.
The newly open-sourced repositories will provide essential infrastructure for DeepSeek’s AI models, expanding upon existing open-source frameworks.
Recent Developments and Popularity
Earlier this week, DeepSeek introduced Native Sparse Attention (NSA), an algorithm designed to enhance training and inference efficiency in AI models with complex contexts, reminds NIXSolutions.
DeepSeek’s chatbot is currently the most popular in China, with 22.2 million daily active users as of January 11, according to Aicpb.com. This surpasses the 16.95 million users of the Chinese social platform Douban.
DeepSeek’s open-source initiative marks a significant step in AI development. We’ll keep you updated on further advancements.