The current balance of power in open models

Newsletter/Podcast · Listen · (gathered 2026-09-23)

• Chinese AI companies have established a clear and growing lead in the open-weight model ecosystem, surpassing American counterparts in capabilities, adoption, and academic influence, posing a significant challenge to U.S. leadership in AI.
• This leadership is evident in metrics like Hugging Face Downloads (China leads 3.2B to 1.6B), performance on benchmarks like the Artificial Analysis Intelligence Index (AAII), and market share on inference platforms like OpenRouter (over 80% Chinese models).
• Chinese open-weight models, such as Z.ai’s GLM-5.3 and Moonshot AI’s Kimi K3, are now foundational in academic research (mentioned in over 40% of AI papers on arXiv) and are being adopted by prominent American companies like DoorDash and Airbnb.
• While distillation contributes to Chinese model success, it only accounts for a 1-2 month reduction in the performance gap, not the full story of their rapid advancement and economic viability.
• To counter this trend and manage the emerging risks of powerful open models, the U.S. must significantly increase investment in domestic open model development to re-establish leadership and avoid reliance on foreign models.