ByteDance Trains 10 Trillion Parameter AI Model Rivaling Anthropic

ByteDance initiates pre-training for a 10T parameter model, three times Kimi K3's size. The project aims to match Anthropic's Mythos standards, signaling intensified global AI competition.

Woofun AI reports that ByteDance has commenced early-stage training for an ultra-large-scale artificial intelligence model with a potential parameter count of 10 trillion. This scale is approximately three times that of Moonshot AI’s largest existing model, Kimi K3. The model is currently undergoing pre-training, a phase typically lasting three to six months, with fine-tuning and release scheduled for a later stage. Industry estimates suggest Anthropic’s advanced Mythos 5 model contains around 8 trillion parameters, while Fable 5 holds approximately 5 trillion. ByteDance aims to reach standards comparable to these state-of-the-art systems, though the final model scale remains undetermined.