In the realm of artificial intelligence, the pace of change is relentless. A mere few months after our spring report, we present fresh insights from the evolving ecosystem of open models as observed from January to August 2026.
Growth in Public Repositories
The Hugging Face hub has seen remarkable growth, with public model repositories increasing from 2.43 to 2.96 million, datasets rising from 711,000 to 1 million, and Spaces expanding from 1.00 to 1.44 million. However, the distribution remains skewed: approximately 85.6% of models have fewer than 200 lifetime downloads, while a mere 1.5% of repositories account for 99.2% of all downloads.
Shifting Frontiers
The traditional trajectory of model development is changing. In 2026, several Chinese labs have bypassed the gradual progression from smaller to larger models. The largest open models from these labs have reached between 754 billion and 2.78 trillion parameters, while American labs have largely remained below 130 billion parameters. Notably, NVIDIA’s Nemotron 3 Ultra peaked at 561 billion parameters in May and June, but most American releases above 100 billion are adaptations of Chinese models.
Adoption Metrics
Analysis of the top 25 model repositories by downloads and likes reveals a stark contrast: only one repository appears on both lists. Models published in 2026 have not yet reached the download top 25, with many established models from 2022 dominating. Downloads reflect operational dependencies, while likes indicate immediate excitement. This disparity highlights the different metrics of engagement within the community.
Licensing and Ecosystem Positioning
Interestingly, the licensing terms of models reveal a trend toward permissiveness. Among 178 Chinese releases above 20 billion parameters, 59% are licensed under Apache 2.0 or MIT, with none carrying non-commercial restrictions. In contrast, American models are often under more restrictive terms. This suggests that the value derived from these models may not stem from licensing revenue but rather from API, cloud services, or hardware positioning.
Qwen’s Dominance
Qwen has emerged as a foundational model in the open ecosystem, with 151,448 derivatives on the Hub, significantly outpacing competitors. Its consistent release schedule, broad coverage across model sizes, and permissive licensing have made it a go-to choice for developers. The community’s engagement with Qwen has fostered a robust ecosystem, further solidifying its position.
The Role of Small Models
Despite the rise of large models, small models remain crucial, accounting for 83% of all-time downloads. The advent of llama.cpp has enabled local inference of larger models, allowing developers to leverage advanced capabilities on consumer hardware. This shift illustrates the evolving landscape where accessibility and performance coalesce.
This article was produced by NeonPulse.today using human and AI-assisted editorial processes, based on publicly available information. Content may be edited for clarity and style.
Original source: huggingface.co
