You will find a number of really interesting graphs in the Stanford

You will find a number of really interesting graphs in the Stanford Institute for Human-Centered Artificial Intelligence (HAI) 2026 AI Index Report. This one is both revealing and a bit misleading. Revealing because Chinese models are indeed "closing the gap" (and then some on certain dimensions), something that seemed to take everyone in the "Western World" by surprise with the February 2025 release of DeepSeek AI DeepSeek-R1, a DeepSeek moment some compared to a Sputnik moment. A frenzy of real innovation ensued from both the US/West AND China. Great. But the misleading part is that this could have been framed as Closed-Source vs Open-Source (or Open-Weights) as the leading US models are closed, proprietary models and the leading Chinese models are all open. And around 41% of downloads on Hugging Face in the past year are from Chinese models, according to their own “State of Open Source” report, with the (large) Qwen family accounting for a cumulative 700+ million downloads. To be clear, this is JUST on Hugging Face. So the proper framing for this graph is both Chinese and Open-Weight: BOTH are critical context. Of course, that also raises the important geopolitical question: why are Chinese companies making their models open while the US frontier labs are closed? Given the cost of developing and training these models, the economics is mysterious, at least in the short term. Longer term, the economic benefits depend as much on business strategy as on the Chinese government's grand strategy. Both may converge if the world starts depending on a steady influx of "free" innovation from China: remember the old adage, if it's free, YOU are the product.