According to "LanDian LatePost," ByteDance is discussing the training of a large language model with parameters exceeding 500 billion. If ultimately launched, its scale will surpass Alibaba's Tongyi Qwen3.8-Max with 240 billion parameters and Moonshot AI's Kimi K3 with 280 billion parameters, becoming the largest known parameter model in China. However, the project is still in the early stage, and it is not yet determined whether it will be officially released.

The report states that the model is planned to be led by Xiang Liang, head of ByteDance's Seed Foundation, and cooperated with Shen Ke, who is responsible for pre-training data of large language models. Around this project, the Seed team is carrying out organizational adjustments, including redefining R&D responsibilities and optimizing resource allocation to support the training of ultra-large-scale models.

According to the information, Xiang Liang and Shen Ke are both key members of ByteDance's artificial intelligence and large model development system. Both of them came from ByteDance's core "search, advertising, and recommendation" technology system. Xiang Liang graduated from the University of Science and Technology of China with a bachelor's degree and earned his doctorate at the Institute of Automation, Chinese Academy of Sciences. He joined ByteDance in 2016 and was previously in charge of the machine learning platform AML team, later serving as the head of the Douyin large model Foundation team. Shen Ke joined ByteDance after graduating from Tsinghua University in 2018 and currently mainly oversees pre-training data-related work for large language models.

As AI manufacturers both domestically and internationally continue to expand the scale of their models and compete in areas such as training data, computing power, and infrastructure, the development of ultra-large-scale models has become an important direction for leading companies to explore general artificial intelligence capabilities. ByteDance's move also reflects that the competition in large models in China is further extending from application innovation to fundamental model capability construction.