ByteDance, the Chinese company that owns TikTok, is reportedly developing a massive new artificial intelligence model with up to 10 trillion parameters. The project shows that ByteDance is increasing its efforts to compete with some of the world’s biggest AI companies.
According to a report by the Financial Times, people familiar with the project said the new model could become as large as, or even larger than, Anthropic’s advanced Mythos AI system.
However, ByteDance’s new model is still in the early stages of development, and its final size and capabilities could change before it is completed.
Model Is Still Being Trained
The new ByteDance AI model is currently in the pre-training stage. This is one of the most important and time-consuming parts of developing a large AI system.
Pre-training can take several months. Developers first train the model using huge amounts of data before moving on to other stages, such as fine-tuning and testing.
The ByteDance project could eventually reach 10 trillion parameters, although this number has not been officially confirmed by the company.
Parameters are the values an AI model learns during training. A model with more parameters can have a larger capacity, but a bigger parameter count does not automatically mean that it will be smarter or perform better.
The quality of an AI model also depends on its training data, architecture, training methods, computing power, and how efficiently the model uses its parameters.
Bigger Than Many Rival Models
If ByteDance’s model reaches 10 trillion parameters, it would be one of the largest AI models currently being developed.
Industry estimates cited by the Financial Times suggest that Anthropic’s Mythos model has around 8 trillion parameters.
Other large AI models are believed to be smaller. Moonshot AI’s Kimi K3 is estimated at around 2.8 trillion parameters, while Meituan’s LongCat-2.0 and DeepSeek’s V4-Pro are estimated at around 1.6 trillion parameters each.
These figures are estimates because many AI companies do not publicly reveal the exact number of parameters in their latest models.
ByteDance Is Spending More on AI
The new project is part of ByteDance’s larger plan to become a major player in advanced artificial intelligence.
The company has a dedicated AI research and development group called Seed. The team works on areas such as AI model training, post-training, inference, memory, learning, and understanding how AI models make decisions.
ByteDance also has teams working on the computer infrastructure needed to train and operate very large AI models.
The company has reportedly invested heavily in AI during the past three years. It has expanded its data centre capacity and hired more AI researchers in China and other countries.
ByteDance’s Seed team is reported to have around 2,000 employees worldwide. The team is led by Wu Yonghui, a former Google DeepMind scientist.
Focus on Its Own AI Technology
ByteDance has also been trying to develop its AI technology more independently.
According to the report, the company has avoided relying heavily on model distillation from competing AI companies for more than a year.
Model distillation is a process in which a smaller AI model is trained to copy or learn useful behaviour from a larger and more powerful model.
Instead, ByteDance is reportedly focusing on developing its own models and training methods.
ByteDance founder Zhang Yiming reportedly believes that the company needs to build its own technology if it wants to eventually compete with or outperform major AI companies.
During an internal meeting recently, Zhang reportedly encouraged the Seed team to focus on achieving world-class AI capabilities over the long term instead of worrying about whether the company is temporarily behind its competitors.
ByteDance Also Expands AI Infrastructure
ByteDance is not only working on AI models. It is also expanding the infrastructure needed to support them.
The company has been growing its Volcano Engine cloud business, which provides cloud and AI services to companies.
ByteDance also has ambitions to develop its own AI chips. Developing its own hardware could help the company reduce its dependence on outside chip suppliers and give it more control over AI training and computing.
Building a model with trillions of parameters requires enormous computing power, large data centres, advanced chips, and efficient software systems.
Release Date Is Still Unknown
There is currently no confirmed release date for ByteDance’s new AI model.
Since the model is still in the pre-training stage, it could take several more months before it reaches later development and testing stages. The final model could also have fewer or more parameters than the current target.
ByteDance has not officially confirmed the reported details about the project. Reuters also said it had not independently verified the information.
If the reported 10-trillion-parameter target is reached, however, the project would highlight ByteDance’s growing ambitions in the global AI race and its efforts to compete with major US and Chinese AI developers.
