ByteDance is developing one of the world’s largest artificial intelligence models in an effort to rival Anthropic’s flagship Mythos model, according to a report by the Financial Times. The Chinese technology giant is reportedly in the early stages of training a frontier AI model that could reach up to 10 trillion parameters, significantly larger than any AI model previously announced by a Chinese company. The initiative reflects ByteDance’s ambition to compete directly with leading U.S. AI labs as the global race for next-generation foundation models accelerates.
The report says the model is currently undergoing pre-training, a computationally intensive phase expected to take three to six months before fine-tuning and eventual release. While parameter count alone does not determine AI capability, a model of this scale would place ByteDance among a small group of companies—including Anthropic, OpenAI, and Google—building frontier AI systems at the highest level. Reuters said it could not independently verify the Financial Times report.
ByteDance Targets Frontier AI Leadership
According to people familiar with the project cited by the Financial Times:
- ByteDance is training a model with as many as 10 trillion parameters.
- The model remains in the early pre-training stage.
- Final model size may change during development.
- The goal is to approach the capabilities of Anthropic’s Mythos model.
Project Snapshot
| Item | Details |
|---|---|
| Company | ByteDance |
| Reported Model Size | Up to 10 trillion parameters |
| Development Stage | Pre-training |
| Estimated Training Time | 3–6 months |
| Target Competitor | Anthropic’s Mythos |
Larger Than Previous Chinese AI Models
The reported model would represent a significant leap for China’s AI industry.
According to the report:
- It would be roughly three times larger than Moonshot AI’s Kimi K3, currently among China’s largest publicly known models.
- Industry estimates suggest Anthropic’s Mythos contains around 8 trillion parameters, though Anthropic has not officially disclosed its architecture.
- Parameter count measures a model’s capacity but does not directly determine intelligence, as training data, architecture, and optimization techniques are equally important.
ByteDance Is Taking an Independent Development Approach
Unlike some AI developers that improve models using model distillation—training on the outputs of larger AI systems—ByteDance is reportedly pursuing independent model development.
The report says the company believes building models from scratch will deliver stronger long-term capabilities, even if it slows short-term progress. Founder Zhang Yiming recently reiterated internally that ByteDance should prioritize “world-leading model capabilities” over catching up quickly through shortcuts.
Strategic Priorities
| Focus Area | ByteDance’s Approach |
|---|---|
| Model Development | Independent training |
| AI Infrastructure | Large-scale data centres |
| Enterprise AI | Volcano Engine cloud platform |
| Long-Term Goal | World-leading AI capabilities |
Massive Investment in AI
Over the past three years, ByteDance has significantly expanded its AI operations.
The company has invested in:
- AI research talent.
- Large-scale data centres.
- Proprietary model development through its Seed research team.
- Enterprise AI services via Volcano Engine.
- Custom AI chip development.
ByteDance’s consumer AI assistant Doubao has become China’s most widely used AI chatbot, while its SeeDance model ranks among the leading AI video-generation systems globally.
China’s AI Race Intensifies
The reported project highlights how Chinese AI companies are rapidly narrowing the gap with leading U.S. developers.
Recent months have seen major model releases from companies including:
- ByteDance.
- Alibaba.
- Moonshot AI.
- Other leading Chinese AI labs.
Industry observers say several Chinese companies are now training models approaching the scale of Anthropic’s frontier systems, with ByteDance currently pursuing the most ambitious effort.
Looking Ahead
If ByteDance successfully completes training, the new model could become one of the largest AI systems ever built, positioning the company as a direct challenger to frontier AI developers such as Anthropic, OpenAI, and Google. The project also reflects China’s growing determination to achieve technological leadership through independently developed foundation models rather than relying on techniques such as model distillation.
Looking ahead, attention will focus on whether ByteDance can translate the model’s enormous scale into meaningful performance gains. While larger parameter counts increase a model’s capacity, real-world capability depends on training quality, architecture, inference efficiency, and safety. If successful, the project could reshape the competitive landscape of global AI and further intensify the race to build the next generation of frontier models.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
