The Numbers Behind Chinese Open-Source Models’ Performance and Adoption

Open-source AI developments have kept The Information’s reporting corps busy this summer, with news on regulation, Meta Platforms’ return to open source, and the rise of Chinese labs, all culminating with Nvidia’s $12.9 billion purchase of the Hugging Face and the potential $40 billion valuation for startup Thinking Machines Lab.

Open source’s rise raises urgent questions about the geopolitics and the relative costs and performance of open and closed models. To better understand the landscape, I’ve been examining open source models as a supply chain, digging into a rich data trove from Hugging Face that I think is under-discussed.