ByteDance Reportedly Pre-Training 10T-Parameter AI Model, Far Exceeding Kimi

rohanpaul_ai · x · 2026-08-07

ByteDance is reportedly pre-training an AI model with up to 10 trillion (10T) parameters, far exceeding the 2.8T parameter size of the recently released Kimi K3, according to the Financial Times.

FT also noted that ByteDance has avoided distilling rival models for over a year, preferring independent model development. If this reported frontier-scale run succeeds, it will demonstrate ByteDance's capability to execute massive pre-training independently without leaning on a rival model as a teacher.

Related event: ByteDance reportedly pretraining 5-10T-param model; Zhang Yiming opposes distillation(19 posts)→

Original post →

More from Models

Models channel →