Chinese AI models from companies like DeepSeek and Z.ai are gaining rapid traction among U.S. businesses, driven by pricing that runs 60 to 90 percent below comparable American systems — a shift that is redrawing the competitive landscape of the global AI industry.
On the developer platform OpenRouter, Chinese models have accounted for more than 30 percent of total traffic every week since early February, peaking at 46 percent in recent months. A year ago, that figure stood at just 11 percent. The surge reflects a pragmatic calculation by startups and enterprises alike: when open-source models deliver comparable performance at a fraction of the cost, the economic case for paying premium prices to OpenAI or Anthropic weakens considerably.
The shift is already reshaping corporate decisions. Lindy, an AI startup, moved all of its traffic from Anthropic's Claude to DeepSeek. CEO Flo Crivello said the switch saves the company millions of dollars annually. OpenRouter employee Justin Summerville confirmed that Chinese open-source models consistently run 60 to 90 percent cheaper than their Western counterparts.
The quality gap, meanwhile, is narrowing. Kyle Chan, a researcher at the Brookings Institution, estimates Chinese models trail their American competitors by six to nine months. That assessment aligns with a May report from the Center for AI Standards and Innovation, a U.S. government agency, which found Chinese AI systems lagging behind leading American models by roughly eight months across cybersecurity, software development, math, science, and abstract reasoning benchmarks.
Z.ai's GLM-5.2, released in June under an MIT license, has become a focal point of the trend. A hands-on comparison by Snowflake across 103 tasks found GLM-5.2 nearly tied with Anthropic's Claude Opus 4.7 after three attempts. The model has quickly built a following among developers who view it as a credible, low-cost alternative.
The pricing dynamic is forcing a reckoning in Silicon Valley. As per-token costs from U.S. providers continue to climb, the question for many companies is no longer whether Chinese models are good enough — but whether they can afford not to use them.
Image source: i.ibb.co