Hugging Face published its biannual 'State of Open Models' report on August 14, 2026, covering observations from January to August 2026. According to the report, public model repositories on the Hugging Face hub grew from 2.43 to 2.96 million over the period, datasets from 711,000 to 1 million, and Spaces from 1.00 to 1.44 million. The distribution remained concentrated: roughly 85.6% of models had fewer than 200 lifetime downloads, and 1.5% of repositories accounted for 99.2% of all downloads. The report states that in almost every month of 2026, the largest open model from a Chinese lab exceeded anything a U.S. lab released, with China's monthly ceiling running between 754 billion and 2.78 trillion parameters while the U.S. ceiling stayed under 130 billion in five of seven months. Exceptions cited include NVIDIA's Nemotron 3 Ultra (561B) and Thinking Machines Lab's Inkling (952B). AMD and NVIDIA were named as the organizations publishing the most new open models, each with more than 200 repositories, with LiquidAI third at around 100. The report also notes that attention does not equal adoption: of the top 25 repositories by downloads and top 25 by likes, only one appeared on both lists, and all-MiniLM-L6-v2 was downloaded 1.55 billion times against 5,156 likes.
- Public model repositories on Hugging Face grew from 2.43 to 2.96 million between January and August 2026
- 85.6% of models have fewer than 200 lifetime downloads; 1.5% of repositories account for 99.2% of downloads
- China's monthly open-model ceiling ran 754B to 2.78 trillion parameters; U.S. stayed under 130B in five of seven months
- AMD and NVIDIA each released more than 200 new open model repositories, the most of any organizations
What it means for you
This is a state-of-the-field snapshot, not a product you can use. Its most practical takeaway: the models people actually rely on day-to-day are small, stable, and often years old, while the giant new models grabbing headlines are rarely what businesses run in production. The gap between what's exciting and what's dependable is wide.
Try this
If you use open models, check whether your pipeline depends on small proven workhorses (like embedding models) versus chasing the newest large release — the report suggests the older, smaller options are usually the safer bet.
Who should care
Developers and technical teams choosing which open models to build on, and anyone tracking where open-source AI is heading.
Skip this if
You use AI through finished products like ChatGPT or Copilot and never touch model selection yourself — none of this changes what you do.
Sources: Hugging Face — read the original