ByteDance is reportedly developing a massive ByteDance AI model that could contain as many as 10 trillion parameters, marking an ambitious new step in China’s growing artificial intelligence race with leading US technology companies. The project is still in its early training phase, but its reported scale could make it one of the largest AI models ever developed.
According to a report by the Financial Times, people familiar with the project said the model could eventually approach or exceed the estimated size of some of the most advanced systems being developed by Western AI companies. However, the reported parameter count remains an early estimate and could change as development continues.
ByteDance Model Still in Early Training
The new ByteDance AI model is currently believed to be in the pre-training stage. This is one of the most demanding phases of AI development, requiring enormous amounts of computing power, data and infrastructure.
Pre-training can take several months, after which developers typically move to additional stages such as fine-tuning, testing and safety evaluation. The final model may therefore look significantly different from the system currently being trained.
A model with 10 trillion parameters would represent an enormous increase in scale compared with several other major AI systems being developed in China. For comparison, industry estimates place Moonshot AI’s Kimi K3 at around 2.8 trillion parameters, while some estimates put Meituan’s LongCat-2.0 and DeepSeek’s V4-Pro at approximately 1.6 trillion parameters.
Bigger Does Not Always Mean Better
Although the reported size of the ByteDance AI model has attracted significant attention, parameter count alone does not determine how capable an AI system will be.
Parameters are essentially numerical values that help a model identify patterns and generate outputs. A larger model can potentially learn more complex relationships, but its performance also depends on the quality of training data, model architecture, optimization techniques and the amount of computing used during training and inference.
This means a smaller, more efficiently designed model can outperform a much larger system in particular tasks.
The reported 10-trillion-parameter target should therefore be viewed primarily as an indication of ByteDance’s ambitions rather than proof that the company has already developed a superior AI system.
ByteDance Expands Its AI Infrastructure
The project is part of ByteDance’s broader investment in artificial intelligence. Its Seed research team works on several areas, including pre-training, post-training, inference, memory, learning and model interpretability.
The company has also expanded its computing infrastructure to support increasingly demanding AI workloads. According to reports, ByteDance has invested heavily in data centres and recruited researchers from around the world.
Its Seed team reportedly has around 2,000 employees across China and overseas. The group is led by Wu Yonghui, a former Google DeepMind scientist.
ByteDance is also developing its Volcano Engine cloud business, which provides computing and AI services to businesses. The company has further expressed interest in developing its own AI chips, potentially reducing its dependence on external hardware suppliers.
A More Independent AI Strategy
Another notable aspect of the ByteDance AI model project is the company’s reported decision to pursue more independent model development.
ByteDance has reportedly avoided relying heavily on model distillation from competing AI systems for more than a year. Distillation is a technique in which a smaller model learns to reproduce or approximate the behaviour of a larger model.
The company’s founder, Zhang Yiming, has reportedly encouraged the Seed team to focus on long-term technological leadership rather than short-term comparisons with competitors.
During a recent internal meeting, Zhang reportedly urged researchers to pursue world-leading AI capabilities even if the company temporarily falls behind rivals.
That approach suggests ByteDance views artificial intelligence as a long-term strategic priority rather than simply another product category.
Growing US-China AI Competition
The development of the ByteDance AI model comes as competition between Chinese and US technology companies continues to intensify.
American companies such as Anthropic and OpenAI have invested heavily in increasingly capable foundation models, while Chinese companies including DeepSeek, Moonshot AI, Meituan and ByteDance are rapidly expanding their own AI research programmes.
The reported scale of ByteDance’s project demonstrates how Chinese technology companies are continuing to invest in large-scale AI despite challenges involving computing resources, advanced chips and international technology restrictions.
However, the eventual impact of the project will depend on whether ByteDance can turn its enormous training effort into a model that delivers meaningful improvements in reasoning, coding, multimodal understanding and other real-world capabilities.
For now, the ByteDance AI model remains a work in progress, and there is no guarantee that a 10-trillion-parameter system will eventually be released publicly.
The company has not officially confirmed the reported specifications, and the information has not been independently verified.
Still, the project highlights the scale of ByteDance’s ambitions. If development proceeds successfully, it could become an important competitor in the global AI market and further intensify the race to build increasingly powerful foundation models.
The significance of the project will not be measured simply by how many parameters it contains. Its real test will be whether ByteDance can combine massive computing power, high-quality data and advanced architecture to create an AI system that delivers capabilities that genuinely move the industry forward.
