ByteDance is discussing training a model exceeding 5 trillion parameters, which
would surpass Alibaba’s Qwen3.8‑Max (2.4 trillion params) and Moonshot AI’s K3
(2.8 trillion) to become the largest-known model in China. The project is at an
early stage and does not guarantee a release. It would be led by Seed Foundation
head Xiang Liang with LLM pretraining-data lead Shen Ke. Seed is reorganizing
its structure and reallocating responsibilities and resources to support the
effort.
