Hugging Face researchers Adina Yakefu and Irene Solaiman report that China's open-source AI ecosystem adopted Mixture-of-Experts architectures as the default following DeepSeek R1's release. The article documents a shift toward multimodal models, smaller parameter sizes, and more permissive licensing, with Apache 2.0 becoming near-standard. Chinese companies also increasingly aligned model releases with domestic hardware and inference frameworks, prioritizing sustainable deployment and cost performance over maximum capability.
No score is assigned. Sources and their independence are shown in the citation chain below.