ByteDance Is Training a 10 Trillion-Parameter Model, Leaning Into Scale to Chase Anthropic

ByteDance is reportedly in the early stages of training a massive AI model with up to 10 trillion parameters—three times larger than the biggest Chinese model released to date—according to three people with knowledge of the matter. The move positions the TikTok parent as the most ambitious Chinese lab yet in the race to match or exceed the top US models from Anthropic.

The model, being trained by ByteDance's Seed team, is currently in pre-training, a phase that typically lasts three to six months, followed by fine-tuning and potential release. The exact parameter count could change, but the reported scale signals a bet on sheer size as a path to frontier capability.

Industry estimates suggest Anthropic's Mythos 5 has around 8 trillion parameters, and its Fable 5 around 5 trillion. While parameter count sets fundamental capacity limits, capability also depends on data quality and training methods—factors that US labs have long used to maintain their edge.