Chinese AI models now handle between 30 and 46 percent of the tokens US companies are sending through public developer platforms. That is according to CNBC’s analysis of OpenRouter traffic, the biggest router of API calls to AI models.

What changed: GLM-5.2, a Chinese model, landed on Vercel in late June and saw 27x growth in daily token volume in its first week. That is the fastest adoption of any model Vercel has tracked all year. Lindy, a major AI startup, moved 100 percent of its workload from Claude to DeepSeek, a Chinese model. One company told me the switch would save them millions annually.

The driver is simple: cost. Chinese open-source models run 60 to 90 percent cheaper than OpenAI or Anthropic. When you have a task that does not need Claude Opus or GPT-5.6, you route it to the model that is good enough and costs a fifth as much.

This is how markets work. You raise the price on something scarce or powerful, and people find an alternative. For months, Anthropic and OpenAI had the market to themselves because there was no alternative. Now there is. The cost gap is too big to ignore.

The uncomfortable part: these Chinese models are good. GLM-5.2 lands within a percentage point of Claude Opus on agentic benchmarks. They are not inferior knockoffs that users begrudgingly tolerate. They are legitimately capable and significantly cheaper.

What happens next is important. Either US labs close the cost gap, or they accept that they will handle premium use cases while Chinese models own the mainstream.

——

Follow: @Ali Demi
Book your free AI clarity call, NOW!
https://buff.ly/TpWy277

——

Sources:
https://www.cnbc.com/2026/07/07/chinese-ai-models-costs-us-openai-anthropic.html
https://restofworld.org/2026/when-americans-choose-chinese-ai/
https://invezz.com/pk/news/2026/07/07/cheap-capable-and-controversial-why-us-companies-cannot-resist-chinese-ai-models/

Repost this. Thanks.