China’s Chatbot “DeepSec” Announces Open Source Model Code Release

China’s AI chatbot, DeepSec, has announced that it will be making its model code available to the public. The company stated on social media platform X that it will open its Source 5 code repository next week, describing this move as a small but sincere step towards greater transparency. The company emphasized that it will share the model code with full transparency.

DeepSec made headlines last month when it released its open-source R1 Regeneration model, which, despite being cheaper to develop, competes with Western systems in terms of performance. This commitment to open-source has set DeepSec apart from most AI firms in China, which, unlike their American counterparts, tend to lean towards closed-source models.

The newly released open-source code will provide the infrastructure necessary to support AI models that DeepSec has already shared publicly, placing it at the forefront of current open-source model frameworks.

This announcement comes after DeepSec’s release of a new algorithm called Native Sparse Attention, which is designed to make long-context training and estimations more efficient.

DeepSec’s user base is growing rapidly, and it has become the most popular chatbot service in China, with over 22 million daily active users. This new chatbot has surpassed Duban, which has 16.95 million users.

**Cultural Approach Over Commercialism, Says Founder**

DeepSec’s founder, Liang Wenfeng, shared in an interview with Chinese media last July that the company does not prioritize monetizing its AI models. He stated that success comes from others following your innovation, and that open-source is more of a cultural attitude than a commercial one. He added that collaborating in this way brings respect.