According to industry statistics and Stanford-related reports, Chinese open-source models now account for 41% of global downloads, surpassing the US for the first time. Meanwhile, the performance gap between top Chinese and US large models has narrowed to about 2.7%.

The domestic large-model ecosystem shows an intense update pace: in August alone, more than 30 domestic open-source projects were updated, with about 1.1 new models released daily on average. Leading vendors no longer blindly compete on parameter scale, focusing instead on computing efficiency and real-world business outcomes.

Industry insiders say that as domestic models rapidly mature in performance and ecosystem, enterprises have richer low-cost options for model selection, and AI commercialization is moving from proof-of-concept to large-scale production.