Beijing (Reuters) - A Chinese startup announced on Friday its decision to make its models' code publicly available, reinforcing its dedication to open-source artificial intelligence. The company disclosed on social media platform X that it will unveil five code repositories next week, describing the initiative as "small but sincere progress" that it will share "with full transparency."
The post emphasized that these fundamental components in their online service have been thoroughly documented, deployed, and tested in real-world scenarios. Last month, DeepSeek introduced its open-source R1 reasoning model, which matched Western systems in performance at a lower development cost.
Distinct from most AI firms in China and their U.S. counterparts, DeepSeek's focus on open-source has set it apart. In a rare interview with a Chinese media outlet last July, the company revealed its stance on not prioritizing monetization of AI models, highlighting the soft power gained through open-source contributions.
According to Liang, the CEO, "Having others follow your innovation gives a great sense of accomplishment." He added that "open source is more of a cultural behavior than a commercial one, and contributing to it earns us respect."
The newly released open-source code will serve as infrastructure to support the AI models previously shared publicly by DeepSeek, expanding on the existing open-source model frameworks.
Following the announcement, DeepSeek recently unveiled a new algorithm, Native Sparse Attention (NSA), aimed at enhancing long-context training and inference efficiency. The company's user base has surged in recent times, with 22.2 million daily active users in China as of January 11, making it the most popular chatbot service, surpassing Douban's user numbers, as reported by Aicpb.com, a Chinese website tracking AI products.