China's AI developers race to trillion-parameter models amid US export controls
Chinese AI developers are accelerating their push into trillion-parameter foundation models, seeking to narrow the gap with US rivals like OpenAI and Anthropic. The US-China tech divide deepened after Anthropic suspended global access to its top-tier models Mythos and Fable following Trump administration export controls. Chinese firms are moving from billion-parameter models to trillion-parameter architectures with million-token contexts and domestic chip adaptation, per Donghai Securities. DeepSeek launched its first trillion-parameter model V4 in late April. Xiaomi and Alibaba also rolled out trillion-parameter models, with Alibaba's Qwen-3-Max-Preview debuting in September.
Key facts
- Chinese AI developers are accelerating into trillion-parameter foundation models.
- US export controls block foreign access to leading US AI software.
- Anthropic suspended global access to Mythos and Fable after Trump administration export controls.
- Chinese firms shift from billion-parameter to trillion-parameter architectures with million-token contexts.
- DeepSeek launched its first trillion-parameter model V4 in late April.
- Xiaomi and Alibaba have rolled out trillion-parameter models.
- Alibaba's Qwen-3-Max-Preview debuted in September.
- Parameters are a primary measure of AI capabilities.
Entities
Institutions
- OpenAI
- Anthropic
- Donghai Securities
- DeepSeek
- Xiaomi
- Alibaba Group Holding
- South China Morning Post
Locations
- China
- United States
- Washington