A new report from Stanford’s Human-Centered AI Institute, cited by Sina Finance, reveals a notable shift in the open-source AI landscape: in September 2025, Alibaba’s Qwen model family overtook Meta’s Llama to become the most downloaded large language model (LLM) family on Hugging Face.
Between August 2024 and August 2025, Chinese open model developers accounted for 17.1% of total Hugging Face downloads, slightly ahead of US developers at 15.8%. Since January 2025, derivative models based on open models from Alibaba and DeepSeek have outnumbered those based on US and European models.
This data point underscores China’s growing influence in the global open-source AI ecosystem. While US companies like Google and Anthropic still lead in API token consumption—Google at 18.8% and Anthropic at 14.7% as of early 2026—their combined share has nearly halved from earlier levels. Meanwhile, Chinese models have captured nearly 30% of token consumption: MoonshotAI from Dark Side of the Moon at 14.5%, DeepSeek at 9.0%, MiniMax at 4.2%, and Qwen at 2.6%.
The trend suggests that open-source distribution is becoming a key battleground for AI influence, with Chinese developers leveraging permissive licensing and strong community engagement to gain traction. It remains to be seen whether this translates into sustained competitive advantage or provokes a response from US firms.
Why it matters
- Chinese open-source models are gaining global adoption, challenging US dominance in AI.
- The shift in download and derivative model activity signals a changing competitive dynamic in the AI ecosystem.
What to watch next
- Whether US AI companies respond with more aggressive open-source strategies or policy measures.
- How derivative model quality and real-world deployment compare between Chinese and US open models.
Sources
- 中国大模型一年狂飙421%，全球AI格局变天|AI_新浪财经_新浪网 (Sina Finance)
Confidence: 75%