Hugging Face report finds Chinese labs setting the pace on frontier-size open AI models

Hugging Face has published its Summer 2026 State of Open Models report, an analysis of the model repositories, datasets and Spaces hosted on its Hub between January and August 2026. Public model repositories grew from 2.43 million to 2.96 million over the period, datasets from 711,000 to 1 million, and Spaces from 1.00 million to 1.44 million. Roughly 85.6 percent of models had fewer than 200 lifetime downloads, and 1.5 percent of repositories accounted for 99.2 percent of all downloads, the report says.

In almost every month of 2026, the largest open-weight model released by a Chinese lab was larger than any model released by an American lab, according to the report. China's monthly ceiling ran between 754 billion and 2.78 trillion parameters, while U.S. models stayed under 130 billion parameters in five of seven months. The exceptions were NVIDIA's Nemotron 3 Ultra at 561 billion in May and June, and Thinking Machines Lab's Inkling at 952 billion.

Chinese labs took two distinct strategies, the report found. Moonshot, MiniMax, Xiaomi and Z.ai publish almost nothing below 70 billion parameters, so a developer's first encounter with them is a model too large to run on a personal machine. Tencent and Alibaba Qwen cover the whole range from under 1 billion upward. Xiaomi, Ant Group and Meituan all cleared a trillion parameters this year.

The two U.S. organisations publishing the most new open models this year are the hardware makers AMD and NVIDIA — each with more than 200 new model repositories, well ahead of the rest of the field, with LiquidAI ranking third at around 100. Some U.S. releases above 100 billion parameters are built on top of Chinese models or leverage artifacts from Chinese labs; Inkling is one example.

The report also found that likes and downloads on the Hub measure different things. Only one repository appeared in both the top 25 by downloads accumulated this year and the top 25 by likes. all-MiniLM-L6-v2, a small sentence-embedding model, was pulled 1.55 billion times in seven months against 5,156 likes; Kimi-K3 was pulled about 60 times per like it received. Likes track excitement around new releases; downloads track what is wired into working pipelines.