Chinese open models take majority share on OpenRouter and Vercel as Washington opens probes

On Vercel, the share of Chinese models rose to 55% in August from 11% in January, CNBC reported on 26 September. Neither figure is a benchmark score. Both measure what developers and companies actually routed their traffic through. That is why they landed harder in Washington than another leaderboard update would have.

Two U.S. House committees are investigating the impact of the adoption surge, CNBC said. The stated concerns are technology competition, security and Beijing's global influence. AI was also on the agenda as U.S. President Donald Trump and Chinese President Xi Jinping met this week.

The most advanced U.S. models still lead most benchmarks, CNBC noted, and U.S. frontier models still attract more overall spending on the same platforms. The gap between usage share and spending share is the interesting part. Cheap models can win volume long before they win revenue.

Price, then capability

Peter Walker, head of insights at OpenRouter, told CNBC that Chinese open source models released this year "can credibly perform in advanced agentic use cases, especially in regards to coding, in a way that was just not true in late 2025." He added that they are "incredibly cost-effective compared to most models from American labs."

Harpreet Arora, head of agentic infrastructure at Vercel, put the mechanism plainly: "Chinese models are becoming capable enough for more tasks at a much lower cost. Once a model meets the quality bar for the job, that price difference becomes compelling." Arora also said companies still want frontier U.S. models for more complicated tasks. That is a useful caveat against reading the token numbers as a wholesale switch.

OpenRouter's data covers companies in the U.S., Europe and what it defines as the "Global South", 82 countries across Central and South America, Africa and Asia. Vercel did not specify the geographical breakdown of its data.

"Anywhere from Lagos to São Paulo to Jakarta where entrepreneurs and governments are looking for cheap, open models, will look first to Chinese AI."

That line comes from Daniel Remler, a senior fellow in the technology and national security program at the Center for a New American Security, a think tank. Remler told CNBC that Chinese AI represents "real economic and security risks for the United States" and that "the ultimate concern is that the integration of Chinese AI models pulls countries into a Chinese technology sphere of influence that hardens into geopolitical alignment."

The numbers behind that worry are concentrated. More than two thirds, 67%, of the tokens used by companies OpenRouter classifies as Global South go to Chinese models. About half of all tokens on OpenRouter are used by companies in the U.S. Remler singled out Southeast Asia for likely uptake, citing close economic and cultural links with China plus growing digital infrastructure.

Washington's two lines of attack

The U.S. has tried to hold its lead by restricting Chinese AI companies from buying the most advanced chips through export controls. CNBC reported that Washington is concerned about those companies accessing Nvidia chips remotely through overseas data centers, and about "distillation", where new models mimic older, more established ones.

Distillation is hard to police, because a model that learns from another model's outputs looks, in a training log, much like a model that learned from anything else. Export controls are hard to police for a related reason. Chips in a data center in a third country are still chips, and the buyer is a contract, not a shipping address.

U.S. labs responded with price cuts rather than announcements about sovereignty. Earlier this week, OpenAI and Anthropic both announced new, cheaper models. Dianne Penn, head of product management, research and labs at Anthropic, told CNBC the company was trying to make its models' answers "more efficient, so it uses less tokens depending on your effort setting."

Read the CNBC piece and the through line is cost per task, not raw capability. Chinese labs including DeepSeek, Z.ai and Alibaba have shipped models with major gains on coding and other agentic work, the kind of work companies pay for by the token and therefore notice. The benchmarks that dominate release coverage measure something else.

None of this settles whether the shift holds. OpenRouter's week of Sept. 14 is a snapshot, and Vercel's August number is a different snapshot from a platform that did not disclose its geography. What both show is that adoption decisions are being made on price at a point in the quality curve that did not exist a year ago. The House investigations will test whether policy can move faster than that.