China: Open-source cooperation underpins performance, cost-effectiveness of China's large AI models.

Storyline Open-source cooperation underpins performance, cost-effectiveness of China's large AI models Cooperation between China's open-source AI firms is helping the country's AI industry grow faster, experts say. A standout example includes DeepSeek V4's usage of a key optimizer that was tested and proven effective by another Chinese AI firm, Kimi. A researcher from Kimi said their optimizer helped double DeepSeek's training speed. So, in his words, open-source mode actually speeds-up innovation. Yulun, researcher, Moonshot AI's Kimi model： "Our optimizer has accelerated the training of the DeepSeek model, which can double the training efficiency. Therefore, I believe that the open-source technology actually encourages companies to make faster iterations of AI." Staff from China's National Supercomputing Network, a national-level computing services platform, say they plan to stick to an open-source philosophy because its a win-win that helps connect domestic chips, computing power and AI models. An Lei, application development director, Super Computing Network： "The Super Computing Network still adheres to the principles of openness, open-source, and win-win cooperation, aiming to facilitate connection between domestic chips and domestic computing power, especially connection of domestic models, and provide such inclusive technical services and infrastructure." The Super Computing Network has currently brought together over 340 online cooperation partners, offering more than 7,300 types of software and applications. [Restrictions: No access Chinese mainland]