ByteDance reportedly training 10-trillion-parameter AI model, potentially exceeding Mythos 5
The Financial Times reported on August 7 that ByteDance is training a 10-trillion-parameter AI model currently in early pretraining, expected to take 3-6 months. LatePost had earlier reported the scale at 5 trillion.
On August 7, the Financial Times, citing people familiar with the matter, reported that ByteDance is training an AI model with up to 10 trillion parameters. The model is still in the early stages, currently undergoing 3 to 6 months of pretraining; if all goes well, it will move to fine-tuning and release. The exact scale will be determined in later stages. LatePost had earlier exclusively reported the scale at 5 trillion+, surpassing Kimi K3's 2.8 trillion and approaching Anthropic Fable 5.
The new model is reportedly led by ByteDance Seed Foundation head Xiang Liang, in collaboration with LLM pretraining data lead Shen Ke. At the August 6 ByteDance All-Hands, CEO Liang Rubo acknowledged that the Doubao model is not particularly strong in AI coding, citing Anthropic's Claude Code as an example. Founder Zhang Yiming told the Seed all-hands that even if Seed's capability temporarily lags behind major domestic competitors, ByteDance will not use distillation to accelerate training.
ByteDance has also split Feishu into two, with the product team folded into Doubao and the sales line assigned to Volcengine. Zhang Yiming stressed that integrating Volcengine, Feishu, and Doubao is meant to build advantages in compute and data.