Chinese developers released 16 new AI models in September, focusing on lightweight and cost-effective solutions. According to BenchLM, eight major Chinese companies launched 20 large language models between July and September, showing a significant increase in activity compared to the first quarter of the year.
While US developers released 59 models from January through September, which is 50% more than the Chinese output, the gap in development speed is narrowing. Chinese firms are increasingly targeting the market for AI agents that require high volumes of data processing tokens.
Alibaba is currently developing a new Qwen model version with 5 to 10 trillion parameters, building on its August release of a 2.4 trillion parameter model. Chinese AI models are reported to be 20–50 times more cost-effective for specific tasks compared to top-tier models from Anthropic.
The industry remains divided on development speed. Anthropic CEO Dario Amodei has called for a slowdown in AI progress, citing security concerns and the risk of model distillation. Conversely, Chinese developers continue to prioritize market needs, though some, such as DeepSeek, are actively researching safe training methods for AI agents.
