Chinese-built AI models now account for more than 30% of weekly token usage by U.S. companies on OpenRouter, a platform that gives developers access to a range of AI models. That share has peaked at 46% in recent weeks, according to CNBC. The average across the previous 12 months was 11%. In the first half of 2025, it was 4.5%.
The shift is driven by two forces converging: Chinese models are closing the capability gap with U.S. frontier systems, and token prices for the most advanced American models are rising.
The Cost Differential
Open-source and open-weight Chinese models run “60% to 90% cheaper” than leading Anthropic and OpenAI models, according to Justin Summerville, who works on data and analytics at OpenRouter, speaking to CNBC.
That pricing gap is producing real migration. AI startup Lindy moved 100% of its traffic from Anthropic’s Claude models to DeepSeek in June. “We did it, and you could see that cost curve go down, like, crash to the ground,” CEO Flo Crivello told CNBC. He said the switch will save Lindy millions of dollars within months. Crivello added in a post on X that switching to DeepSeek V4 increased performance on many core use cases.
Zhipu AI’s GLM 5.2 saw the fastest adoption of any model tracked by Vercel in 2026. “In its first full week after launch, daily token volume grew about 27x and the number of customers using it grew about 80x,” Harpreet Arora, head of agentic infrastructure at Vercel, told CNBC.
Capability, Not Just Price
The adoption is not purely a cost arbitrage. Kyle Chan, fellow in the John L. Thornton China Center at Brookings, estimated that Chinese models are currently “six to nine months” behind top U.S. rivals, but operate “close to the top American frontier models” at “a fraction of the cost,” according to CNBC.
GLM 5.2 landed within a percentage point of Anthropic’s Opus 4.8 on one closely watched agentic benchmark, at roughly a fifth of the cost. Some researchers have said GLM 5.2 performs on par with top U.S. labs on certain cyber benchmarks, CNBC reported.
On LaunchLemonade, an AI agent platform for regulated industries, GLM 5.2 is now in the top five models despite Claude and ChatGPT still dominating overall usage. “Chinese models like Z.ai and Qwen are becoming options for companies [as] they offer an attractive combination of performance and cost for specific workloads,” CEO Cien Solon told CNBC.
The Routing Logic
“Price is doing the work here,” Vercel’s Arora said. “When a task doesn’t need the best model, teams are beginning to route it to the cheapest one that’s good enough, and the recent wave of models coming out of China is winning that trade.”
This aligns with the broader shift from what the industry has called “tokenmaxxing” (maximizing AI token usage regardless of cost) to intelligent model routing based on task complexity. For always-on agent deployments that generate thousands of API calls per day, the difference between $15 per million output tokens and $3 per million compounds fast.
Yacine Jernite, head of machine learning at Hugging Face, flagged a structural concern: “There is a real risk that users get stuck having to choose between performant but expensive US proprietary models whose price and accessibility can quickly fluctuate, or using Chinese models as the only feasible alternative whenever they want to control costs or own their AI stack.”
The geopolitical dimension adds friction. At the end of June, OpenAI limited the rollout of new models at the U.S. government’s request. Export controls on Anthropic’s Fable and Mythos models were lifted after a standoff with the Trump administration, according to CNBC. The more U.S. policy restricts access to American models, the stronger the economic case for Chinese alternatives becomes.