China's private-sector large models are sweeping Southeast Asia, Africa and the West with multilingual coverage, lower cost and open ecosystems, Knews reported on October 22. In Southeast Asia, where more than 1,200 languages are spoken and Western models' pretraining is over 80% standard English, Alibaba's Qwen and others were pretrained on hundreds of local languages, avoiding the "water-soil incompatibility" of U.S. models.
Aggregator OpenRouter estimates that in the week of October 5-11 global large-model calls totaled 172 trillion tokens, of which Chinese models accounted for 86.74 trillion — 4.5 times the U.S. figure of 18.97 trillion — with multiple Chinese heads long dominating the token-call leaderboard. From DeepSeek and Qwen to Kimi K3, MiniMax M3, Tencent Hunyuan, Zhipu GLM, Step 5 Preview, Doubao and ERNIE, Chinese models are rising as a cluster.
In Africa, tests found Chinese models more accurate and cheaper, and open weights let countries customize with their own data. In Europe and the U.S., multinationals treat Chinese models as cost-cutting tools. The open-source ecosystem is reconstructing the global AI map and broadening access to innovation dividends.
Industry data confirm a new competitive landscape: Chinese models lead not just on stable tech output and price-performance but on an open ecosystem that makes advanced AI reachable for startups and emerging markets alike.
Source: Knews / NetEase
