# Open Model Adoption Signals: OpenRouter and Hugging Face Two public data sources used to measure how much open models, and Chinese open models in particular, are used in practice. ## OpenRouter token volume OpenRouter is an API aggregator. Developers send requests to one endpoint and OpenRouter routes them to hundreds of models across many providers. Because it sits in the middle, it can publish what share of total tokens goes to each model and provider. - **What it measures:** real usage, weighted by volume, across a broad developer base. - **Bias:** over-represents startups, hobbyists and cost-sensitive developers who shop between models. Large enterprises tend to go direct to a cloud or a lab, so their usage is under-counted. Treat it as a strong signal of developer preference, not of total market share. ## Hugging Face downloads Hugging Face is the main hosting site for model weights. Download counts show how often a model family's files are pulled, including the many derivative models people fine-tune and re-upload. - **What it measures:** developer interest and the base models people build on. - **Bias:** a download is not a production deployment, and automated pipelines inflate counts. ## Reading them together OpenRouter shows what is being run. Hugging Face shows what is being built on. When both move the same way, as with Qwen's rise past a billion downloads alongside Chinese providers' growing token share, the adoption signal is robust. Related: [[Open Weights]], [[Chinese Frontier Labs]], [[Token Economics of Inference]]