State of Open Models: Summer 2026 Observations
Hugging Face’s Summer 2026 report tracks open AI model trends from January to August, highlighting shifts in model sizes, licensing, and geographic contributions to the ecosystem.
Public repositories on the Hugging Face Hub grew from 2.43 million to 2.96 million models and from 711,000 to 1 million datasets between January and August 2026, while usage remained concentrated: 85.6% of models had fewer than 200 downloads, and 1.5% of repositories accounted for 99.2% of all downloads. The report notes that most activity still centers on smaller, established models rather than recent releases, reflecting long-term infrastructure dependencies in AI pipelines.
Chinese labs increasingly dominate the largest open models, with monthly releases often exceeding 754 billion parameters, while American labs’ largest models stayed below 130 billion parameters in five of seven months. Labs like Moonshot and MiniMax focus on frontier-scale models, whereas Qwen and Alibaba release models across the full size spectrum, from under 1 billion to trillions of parameters, catering to diverse developer needs.
Hardware vendors such as AMD and NVIDIA led new model releases in 2026, with over 200 repositories each, as open models became a strategy to demonstrate chip compatibility. Meanwhile, Google, Meta, and IBM Granite contributed fewer new models compared to previous years, signaling a shift in open-source leadership toward infrastructure-focused companies rather than traditional model labs.
Frontier Chinese models are downloaded far more frequently than American counterparts, with Moonshot’s releases totaling 37 million downloads and Qwen’s family reaching 2.045 billion downloads in 2026. Licensing also differs: 59% of large Chinese models used Apache 2.0 or MIT licenses, compared to 29% of equivalent American models, which often relied on custom terms or unspecified licenses.