ByteDance Trains AI Model With Up to 10 Trillion Parameters in Early Pre-training Phase

ByteDance is training an artificial intelligence model with up to 10 trillion parameters. The model is currently in the pre-training phase, which typically takes three to six months, after which it will enter fine-tuning and eventually be released.

This parameter scale far exceeds existing models in China. Moonshot AI's Kimi K3 has 2.8 trillion parameters, and ByteDance's new model is more than three times its size. Previously, Meituan's LongCat-2.0 and DeepSeek's V4-Pro led China's domestic AI industry with 1.6 trillion parameters, but Kimi K3 has since moved ahead. In terms of parameter count, industry estimates place Anthropic's most advanced Mythos 5 at about 8 trillion parameters and Fable 5 at about 5 trillion; ByteDance's model approaches these systems in scale.

Parameters are numerical settings that a model learns from data to identify patterns, generate answers, and perform tasks. They are generally viewed as a rough indicator of model size, though actual capability also depends on other factors including data quality and training methods. Chinese technology companies are continually compressing the iteration and release cycles of large models to adapt to intense competition in the global AI arena.