Skip to main content

ByteDance Eyes 5 Trillion Parameter Model, Could Top China's Biggest

ByteDance is quietly laying the groundwork for what could become China's largest AI model yet. According to a report by LanDian LatePost, the tech giant is discussing the training of a large language model with more than 5 trillion parameters. If it moves forward, the model would dwarf the current record holders—Alibaba's Tongyi Qwen3.8-Max with 240 billion parameters and Moonshot AI's Kimi K3 with 280 billion parameters—making it the biggest known model in the country.

But don't pop the champagne just yet. The project is still in its infancy, and there's no guarantee it will ever see the light of day. Sources familiar with the matter say the initiative is being spearheaded by Xiang Liang, the head of ByteDance's Seed Foundation, with Shen Ke, who oversees pre-training data for large language models, playing a key role. The Seed team is already undergoing organizational tweaks—redefining R&D responsibilities and reallocating resources—to support the ambitious undertaking.

Both Xiang and Shen are veterans of ByteDance's core "search, advertising, and recommendation" technology system. Xiang, who holds a bachelor's from the University of Science and Technology of China and a doctorate from the Institute of Automation at the Chinese Academy of Sciences, joined ByteDance in 2016. He previously led the machine learning platform AML team and later headed the Douyin large model Foundation team. Shen, a Tsinghua University graduate who joined in 2018, currently focuses on pre-training data for large language models.

This move comes as AI players worldwide race to scale up their models, competing fiercely on training data, computing power, and infrastructure. For ByteDance, diving into ultra-large-scale models signals a strategic bet on foundational capabilities rather than just application-level innovation. It's a clear sign that the battle for AI supremacy in China is heating up—and the next frontier might be who can build the biggest brain.

Key Points

  • ByteDance is discussing a model with over 5 trillion parameters, which would be China's largest.
  • The project is led by Xiang Liang and Shen Ke, both from ByteDance's search and recommendation background.
  • The model is still in early stages, with no confirmed release.
  • This reflects a broader trend of Chinese tech giants investing in foundational model capabilities.