Report: ByteDance Pre-training Massive Model Nearing 10T Parameters

zijing_wu · x · 2026-08-07

According to Zijing Wu, ByteDance is targeting a mega AI model nearing 10 trillion (10T) parameters in the early stage of pre-training, aiming to compete with Anthropic's estimated 8T parameter Mythos model.

Additionally, multiple Chinese labs are training models around 5T parameters (e.g., Fable est. 5T), while Seed has been pursuing a no-distillation approach for over a year.

Related event: Report: ByteDance Preps 5 to 10 Trillion Parameter AI Model(4 posts)→

Original post →

More from Models

Models channel →