Chinese AI companies have established themselves as the clear leaders in the open-weight model ecosystem. According to recent data, Chinese models have seen a massive surge in adoption, with Hugging Face downloads for Chinese open-weight models reaching approximately 3.2 billion—twice the total for American models. This growth is particularly evident on inference platforms like OpenRouter, where the market share of Chinese models has risen from 70% in September 2025 to over 80% today.
China Emerges as Leader in Open-Weight AI Ecosystem
In terms of capability benchmarks, such as the Artificial Analysis Intelligence Index (AAII), Chinese models currently hold a clear lead. As of mid-September 2026, top Chinese models like Z.ai’s GLM-5.3 and Moonshot AI’s Kimi K3 significantly outscore the leading American models, such as Thinking Machines’ Inkling. The performance gap between Chinese open-weight models and frontier closed models is estimated to be approximately 2 to 5 months.
This leadership extends into academic research. A scan of recent papers on arXiv shows that while American models were dominant in early 2023, the landscape has shifted. As of September 2026, mentions of any open-weight model in top machine learning research categories account for 50% of papers, with Chinese models being mentioned in 40% of cases, surpassing the 30% share held by U.S. models.
The rapid advancement of Chinese labs is attributed to several factors, including faster release cycles and a strategic focus on specific high-demand tasks like agentic coding. While the role of knowledge distillation—a technique used to transfer capabilities from large models to smaller ones—is a significant factor in narrowing the gap with American frontier models, researchers suggest it only accounts for an estimated 1-2 month difference in the performance gap.
Sources
- The current balance of power in open models （Hacker News Frontpage, 2026-09-22）