DeepSeek’s V4.1-Flash-1017 model topped global AI usage in the week of October 19 to 25, processing 33.6 trillion tokens through OpenRouter, up 801% on the previous week, according to Chinese media outlet Caixin, as reported by Aju Press on October 26. Total usage across all large language models on OpenRouter reached 188 trillion tokens for the week, up 14.63% week on week. Chinese models together processed 93.54 trillion tokens against 22.47 trillion for United States models, roughly 4.2 times as much, marking the 26th consecutive week that Chinese models have led American ones. The figures cover API traffic through OpenRouter only, not all AI usage worldwide.
A Clean Sweep For Chinese Models
All five of the week’s most-used models were Chinese. Behind DeepSeek’s V4.1-Flash-1017 in first place, DeepSeek’s V4.1 Flash took second place with 18 trillion tokens. MiniMax’s M3.1 Flash ranked third with 13.7 trillion tokens, followed by GLM 5.3 Flash in fourth with 13.2 trillion tokens and Tencent’s Hy4 Preview in fifth with 8.97 trillion tokens. Chinese models accounted for 49.8% of total OpenRouter usage for the week, while United States models represented 11.9%. The surge for V4.1-Flash-1017 follows DeepSeek’s V4.1-Flash-1017 release earlier this month, which paired a top DeepSWE score with an open search API.
What The Numbers Do And Do Not Show
OpenRouter is a platform that connects developers and businesses to many AI models through one API, and its figures count input and output tokens generated through that route. Models used directly, without going through OpenRouter, are not included, so the ranking measures developer API traffic rather than total global AI use. Within that frame, the pattern is striking: Chinese models grew 18.1% week on week while United States models grew 29.4%, yet the gap in absolute volume widened because the Chinese base was already far larger. Aju Press attributes the high usage of Chinese models to a combination of competitive pricing, performance on specific tasks, and their expanding place in the developer ecosystem, where engineers compare models on performance and cost before choosing one for coding tools or AI agents.
Why It Matters
Usage through a developer gateway is a leading indicator of where working software is actually being built, and this week’s numbers suggest that gravity is shifting. Developers routing work through OpenRouter are choosing Chinese open models in volume, not just testing them, which strengthens the feedback loop of real-world use, bug reports and improvements for those models. The result also lands weeks after Beijing’s approval for DeepSeek and MiniMax to deploy models globally, tying commercial momentum to a policy backdrop that is actively encouraging Chinese models to compete abroad. For American labs, the challenge is no longer only benchmark scores: it is holding developer share against cheaper, capable alternatives that are already handling nearly half of the traffic on one of the main model marketplaces.
The question now is whether the surge holds beyond a launch week. V4.1-Flash-1017’s jump coincided with its recent release, and weekly API traffic can swing sharply when developers test a new model before settling on it for production work. If the model keeps its lead in the coming weeks, the story becomes structural rather than a spike; if it fades, the week’s headline will mark a moment of peak curiosity. Either way, OpenRouter’s weekly counts have become one of the few public windows into which models developers actually run, and the next few releases will test whether the Chinese lead keeps widening.
Background
DeepSeek, the Hangzhou-based lab behind the week’s most-used model, broke into global attention as an open-model challenger whose efficient, low-cost models forced larger rivals to cut prices, as described in DeepSeek’s open-model challenger story. Its V4.1 family has since become a staple of the developer API market, and this week’s usage figures show that position hardening rather than fading.
Source: Aju Press: DeepSeek leads global usage
0 Comments