Parameter counts are an imperfect measure of AI capability, but they still tell you something about ambition. And ByteDance’s latest bet is hard to ignore. According to Reuters, citing a Financial Times report, the TikTok parent is training a model with as many as 10 trillion parameters, a scale that would put it in the same range as Anthropic’s Mythos system, widely considered one of the most advanced AI models currently in existence.
To understand how significant that number is, consider the context. Moonshot AI’s Kimi K3, currently one of the largest models out of China, sits at 2.8 trillion parameters. Before that, Meituan’s LongCat-2.0 and DeepSeek’s V4-Pro led Chinese AI development at around 1.6 trillion total parameters. ByteDance’s reported model would be more than three times the size of Kimi K3, and roughly six times larger than where China’s frontier sat just months ago.
Direct comparisons with U.S. systems are complicated because Anthropic and OpenAI don’t publish parameter counts for models like Mythos, Fable, or GPT-5.5. But industry estimates cited by the FT suggest Anthropic’s Mythos 5 sits at around 8 trillion parameters, and Fable 5 at roughly 5 trillion. That puts ByteDance’s target squarely in the top tier of what anyone, anywhere, is building right now.
The model is currently in pre-training, a phase that typically runs three to six months before fine-tuning and eventual release. So this is not a product announcement. ByteDance did not respond to requests for comment, and Reuters was unable to independently verify the report. Still, the direction is clear.
This matters beyond the headline number. Chinese AI labs have been aggressively compressing their release cycles, trying to close the gap with U.S. competitors while keeping inference costs manageable. The pressure is real: U.S. export controls on advanced chips have forced Chinese companies to find efficiency gains that their American counterparts don’t have to prioritize in the same way. Building at this scale under those constraints would be a meaningful technical achievement, not just a marketing exercise.
ByteDance already has Doubao, one of the most widely used AI applications in China. A frontier-scale model sitting behind that product, or made available to enterprise customers, would change what ByteDance can offer and how it competes with Alibaba, Baidu, and the newer wave of Chinese AI startups that have drawn significant investor attention over the past year. The race isn’t slowing down.
