On September 17, China Telecom AI Technology Co., Ltd. made an official announcement regarding the launch of its cutting-edge XingChen large model, Xing4.0-29B-A4B. Tailored specifically for the agent era, this lightweight code agent large model boasts robust capabilities in comprehending intricate tasks, strategic planning, tool utilization, and autonomous execution. It excels in processing extended contexts, making it a versatile solution for code development, data analysis, office automation, and a variety of life services.

Leveraging a Mixture of Experts (MOE) architecture, the model comprises a total of 29 billion parameters, with only 4 billion actively engaged at any given time. This design ensures both efficiency and scalability. Notably, it stands as China's inaugural large model with hundreds of billions of parameters, trained entirely on domestically-sourced computing power and frameworks. Furthermore, it has undergone extensive optimization to excel in complex engineering tasks, marking a significant milestone in the country's AI development landscape.