More public repositories, highly concentrated use
On August 14, Hugging Face published an analysis of Hub activity during the first seven months of 2026. Public model repositories rose from 2.43 million to 2.96 million, yet 85.6% had fewer than 200 lifetime downloads and 1.5% of repositories accounted for 99.2% of downloads. Only one repository appeared in both the top 25 for in-window downloads and top 25 for likes, illustrating the gap between attention and pipeline dependence.
The report counts 151,448 Qwen derivatives—2.6× Meta's total footprint. Among repositories declaring parameter count, models below 1B accounted for 83% of all-time downloads, while models above 100B accounted for 1%. Repositories declaring GGUF grew 464%, LeRobot 194% and Apple MLX 148%, suggesting that runtime, robotics and local-deployment layers are expanding faster than the modeling core.
The Hub is not the entire market
These are Hugging Face downloads, likes and derivative repositories. They exclude API traffic, private deployment and other distribution channels; a download is neither a unique user nor a quality score. “Open source” must also be distinguished from weights under varying licenses. The report is still useful because it reveals adoption and execution patterns that launch benchmarks alone miss.