Chinese models accounted for 41% of downloads on Hugging Face in the platform’s Spring 2026 ecosystem audit, surpassing the US. (State of Open Source on Hugging Face: Spring 2026) The same report shows independent developers have edged past industry contributors in share of downloads, while the mean size of downloaded open models rose from 827M parameters in 2023 to 20.8B in 2025, though the median stayed near 400M (State of Open Source on Hugging Face: Spring 2026). For teams building fine-tune pipelines and procurement shortlists, the practical implication is to look beyond US frontier labs. The downstream ecosystem includes a much broader set of contributors.
China Accounts for 41% of Downloads in the Report
China’s 41% share in the Spring 2026 report is a plurality of Hugging Face downloads. Release activity grew alongside download share: Baidu went from zero Hub uploads in 2024 to more than 100 in 2025, while ByteDance and Tencent each increased their release counts by eight to nine times over the same period (State of Open Source on Hugging Face: Spring 2026). A companion arXiv study records an earlier turning point in its 2025 data: Chinese model developers surpassed their US counterparts in download share for the first time. It also describes declining US dominance as Chinese industry and unaffiliated developers gained share (Economies of Open Intelligence: Tracing Power & Participation in the Model Ecosystem). The paper and the later blog report cover different snapshots; their percentages should stay attached to those periods.
Industry Share Falls From 70% to 37%; Independents Reach 39%
Industry-affiliated developers accounted for roughly 70% of downloads before 2022; by 2025 that share had fallen to roughly 37%. Independent or unaffiliated developers rose from 17% to 39%, at times accounting for more than half of total usage (State of Open Source on Hugging Face: Spring 2026). The same arXiv study notes that open-weights models surpassed truly open-source models for the first time in 2025 (Economies of Open Intelligence: Tracing Power & Participation in the Model Ecosystem), a distinction that matters for licensing and modification rights even when download counts look similar.
A later release illustrates why the licensing distinction needs care. Zhipu’s GLM-5.2 model card on Hugging Face lists the MIT license under the zai-org organization (GLM-5.2 Model Card and Weights (zai-org on Hugging Face)). That licensing statement does not by itself establish access to the training data or the ability to reproduce training. The companion study finds that Open Use licenses declined as more restrictive licensing and access conditions became more common (Economies of Open Intelligence: Tracing Power & Participation in the Model Ecosystem). For a procurement decision, I’d inspect the particular zai-org release and its terms rather than infer rights from the word “open.” For related coverage, see Zhipu Open-Sources GLM-5.2 Under MIT While Anthropic Tightens Model Access and Zhipu Ships GLM-5.2 With 1M Context and MIT Weights, but Zero Benchmarks at Launch.
Small-Model Downloads and the Gap Between Mean and Median
Comparing the median downloads among the top-10 models in each size category, 1-9B parameter models are downloaded roughly four times more often than models above 100B parameters. More than 1.4 billion of roughly 2 billion total downloads come from the 1-9B range (ATOM Project: Relative Adoption Metric). Automated systems and CI pipelines also inflate small-model counts. Treat the ratio as a download comparison, with those automated uses included (ATOM Project: Relative Adoption Metric).
The mean and median also point to different parts of the size distribution. The mean downloaded model grew from 827M parameters in 2023 to 20.8B in 2025, while the median increased only marginally from 326M to 406M (State of Open Source on Hugging Face: Spring 2026). Larger models pull up the mean while the downloaded-model median remains small. Download records do not establish the size distribution of production deployments.
Qwen’s Derivative Moat: 113,000+ and Counting
At the report’s snapshot, Alibaba’s Qwen family had more than 113,000 derivative models and over 200,000 tagged models total on the Hub. As an organization, Alibaba has more derivative models than Google and Meta combined (State of Open Source on Hugging Face: Spring 2026). For procurement, that activity is a reason to investigate the family’s adapters, quantization configs, and deployment scripts. The count can help you find an active ecosystem; you still need to check whether its artifacts fit your model and workload.
Download Concentration and the Short Engagement Window
A broader set of contributors does not mean downloads are evenly distributed. The top 200 most downloaded models (0.01% of all models) comprise 49.6% of all downloads. Half of all models on the platform have fewer than 200 total downloads (State of Open Source on Hugging Face: Spring 2026). The average model’s active engagement period after release is approximately six weeks (State of Open Source on Hugging Face: Spring 2026). This is a mean engagement duration, not a deadline or a measured half-life for every upload.
What This Means for Western Procurement and Fine-Tune Pipelines
A “frontier lab” shortlist (OpenAI, Google, Meta, Anthropic) leaves out much of the open-weights ecosystem. Procurement teams should check the families that downstream developers are actually adapting. Chinese organizations, particularly Alibaba via Qwen and Baidu via its 2025 release surge, now drive a plurality of downstream activity (State of Open Source on Hugging Face: Spring 2026). The label “open” also says little about the distribution of attention. With the top 200 models capturing half of all downloads, raw volume gives an incomplete picture of the ecosystem (State of Open Source on Hugging Face: Spring 2026).
Derivative count and sustained engagement are useful checks alongside downloads. A model family with 113,000+ derivatives gives you more artifacts to inspect, but that does not make every checkpoint a safe integration choice. Check the specific adapter, license, and deployment support your application will need.
Join the discussion
Share a useful perspective or ask a question about this article.