Chinese AI models grab 46% of US token share as costs plunge 90%

US companies now route more than a third of their AI traffic to Chinese models, as a 60% to 90% cost advantage narrows the performance gap with leading American rivals.

"Price is doing the work here," Harpreet Arora, head of agentic infrastructure at Vercel, said. "When a task doesn't need the best model, teams are beginning to route it to the cheapest one that's good enough."

The share of tokens used by US companies on Chinese AI models via OpenRouter has stayed above 30% each week since Feb. 8, peaking at 46%. That compares with an 11% average over the prior 12 months and just 4.5% in the first half of 2025. Chinese open-source models cost 60% to 90% less than leading Anthropic and OpenAI systems, according to Justin Summerville, who oversees data and analytics at OpenRouter.

The shift threatens the pricing power of US frontier labs at a moment when token prices for their most advanced models are rising. OpenAI and Anthropic have been approached for comment. The trend also complicates US efforts to regulate AI exports — Washington's 18-day shutdown of Anthropic's Fable 5 and Mythos 5 models in June was followed within a day by Chinese lab Z.ai releasing its GLM 5.2 model as open source.

DeepSeek and GLM 5.2 Lead the Charge

Chinese models have moved from curiosity to production workhorse in months. AI startup Lindy moved 100% of its traffic from Anthropic's Claude models to DeepSeek in June, a decision CEO Flo Crivello said will save the company millions of dollars. "You could see that cost curve go down, like, crash to the ground," Crivello told CNBC.

Z.ai's GLM 5.2, released in late June, saw the fastest adoption of any model tracked by Vercel in 2026. Daily token volume grew about 27 times in its first full week, and the number of customers using it grew about 80 times, Arora said. The model landed within a percentage point of Anthropic's Opus 4.8 on one closely watched agentic benchmark at roughly a fifth of the cost.

DeepSeek's share of gateway tokens climbed between May and June on Vercel's platform, which lets developers deploy and run applications. On OpenRouter's rankings, GLM 5.2 now sits at number five by usage.

Performance Gap Narrows to Months

Chinese AI models now operate "close to the top American frontier models" while costing a fraction of the price, said Kyle Chan, a fellow at the Brookings Institution's John L. Thornton China Center. He estimates they trail US rivals by six to nine months.

"The new open source models are performing well and prove capable for all but the most complex LLM tasks," OpenRouter's Summerville said. On LaunchLemonade, an AI agent platform for regulated industries, GLM 5.2 is now in the top five models, CEO Cien Solon said.

The capability gains come as US labs face rising costs. OpenAI and Anthropic have raised token prices on their most advanced systems, pushing cost-conscious enterprises to explore alternatives. "We're seeing companies increasingly motivated to turn to cheaper AI stacks they can control and adapt themselves," Yacine Jernite, head of machine learning at Hugging Face, said.

What It Means for Investors

The adoption shift pressures the valuation premium that US frontier labs and their backers have commanded. If Chinese models continue closing the performance gap while maintaining a 60% to 90% cost advantage, the addressable market for premium US AI systems could shrink to only the most demanding workloads — complex reasoning, high-stakes coding, and regulated industries where data sovereignty concerns block Chinese alternatives. For enterprises, the trend unlocks material cost savings: Lindy's migration alone will save millions within months, according to its CEO. The question is whether US labs can cut prices fast enough to retain market share, or whether the open-source Chinese ecosystem becomes the default choice for the vast middle of the AI market.

This article is for informational purposes only and does not constitute investment advice.