ByteDance is training a massive AI model with up to 10 trillion parameters, potentially outstripping Anthropic's advanced Mythos system. The project, still in early stages, underscores China's push to lead in AI innovation.
TL;DR
- ByteDance's new AI model could be the world's largest, with up to 10 trillion parameters, rivaling Anthropic's Mythos.
- The model is currently in pre-training, with its final size to be decided later, according to the Financial Times.
- ByteDance's focus on long-term innovation highlights China's ambition to surpass U.S. AI leaders.
What happened
ByteDance, the parent company of TikTok, is reportedly developing an AI model with up to 10 trillion parameters, making it about three times larger than Moonshot's Kimi K3, currently China's largest released AI model. The project is still in its early stages, according to the Financial Times. The model is currently in pre-training, a process that usually lasts three to six months. Its final size will be decided later, before potential fine-tuning and release if development progresses successfully.
ByteDance has quietly advanced its AI efforts, keeping most models private compared with other Chinese rivals. Its SeeDance video generation model is considered one of the world's most advanced, while its consumer AI assistant Doubao leads China's market with 324 million monthly active users.
Why it matters
ByteDance's new AI model highlights China's goal of moving beyond catching up with U.S. AI leaders to potentially surpass them. Recent Chinese models from companies like Moonshot and Alibaba Group Holding Ltd's latest Qwen3.8-Max have performed strongly on benchmarks, trailing only Anthropic's Fable 5 in some areas.
Anthropic's most advanced model, Mythos 5, remains restricted to approved organizations following temporary security-related limits. Several Chinese AI labs are reportedly developing extremely large models comparable to Fable 5, with ByteDance leading the race to build the biggest and most ambitious system.
Key facts
- ByteDance's new AI model has up to 10 trillion parameters, potentially making it the world's largest.
- The model is currently in pre-training, a process that usually lasts three to six months.
- ByteDance's SeeDance video generation model is considered one of the world's most advanced.
- ByteDance's consumer AI assistant Doubao has 324 million monthly active users in China.
- Anthropic's Mythos 5 model is estimated to have around 8 trillion parameters, while Fable 5 may have about 5 trillion parameters.
- ByteDance founder Zhang Yiming has urged the company's AI team to focus on building a world-class model through independent development rather than chasing short-term competition.
Context
ByteDance's development of a massive AI model underscores the intensifying competition in the global AI landscape. As Chinese companies like ByteDance, Moonshot, and Alibaba Group Holding Ltd push the boundaries of AI innovation, they are increasingly challenging the dominance of U.S. AI leaders like Anthropic. This race to build larger and more advanced AI models is not just about technological superiority but also about strategic influence and market leadership in the rapidly evolving AI industry.
