Chinese AI models are gaining traction with US companies as costs rise

Chinese-developed AI models are gaining traction with US companies as they narrow the performance gap with leading American competitors while remaining much cheaper to use.

Recently released models from Chinese companies, including DeepSeek and Z.ai, are seen by many as serious competition compared to leading edge systems from the likes of Anthropic and OpenAI. That advance in capability comes as prices for more advanced models are rising at many US AI labs, leaving companies facing unexpectedly high costs associated with deploying the technology.

The share of tokens used by US companies in Chinese AI models via OpenRouter — a platform that enables developers to access a range of AI models — has remained above 30% each week since Feb. 8, and that number rises to 46%. The average over the last 12 months was just 11%, falling to 4.5% in the first half of 2025.

The rise of China’s open-source and open-source models comes as US officials look to increasingly rein in their powerful AI models and consider how to stop the rapid adoption of alternatives from overseas.

In late June, OpenAI said it would slow down the release of a new set of models at the government’s request. Export controls on Anthropic’s Mythos and Fable models were also lifted that month, after tensions between the Trump administration and the company.

“Chinese AI models are especially attractive to American companies now that AI is so expensive,” Kyle Chan, a fellow at the John L. Thornton China Center at the Brookings think tank, told CNBC. “Where American companies previously prioritized AI adoption regardless of the model, they are now becoming more cost-effective.”

As companies look to deploy AI models to create new products and bring internal efficiencies, developers are increasingly experimenting with cheap open sources and open weight models, which are the most powerful features of Chinese companies.

Open source and open source models make different parts of an AI model available for developers to test, use and sometimes modify. They are different from closed systems, such as many of the flagship models produced by OpenAI, Anthropic and Googlewhere the code and inner workings remain proprietary.

In June, AI startup Lindy moved 100% of its traffic from Claude’s Anthropic models to DeepSeek, a Chinese company that burst onto the scene with a bombshell release in early 2025 and launched a new model in April.

“We did it, and you can see that the cost curve is going down, like, it’s crashing down,” CEO Flo Crivello told CNBC. He said this decision will save Lindy millions of dollars within months.

DeepSeek saw its share of gate tokens increase between May and June on Vercel, a platform that allows developers to deploy and run applications and websites.

Z.ai’s GLM 5.2 was released to much fanfare in June and saw the fastest adoption of any model that Vercel is pursuing in 2026, Harpreet Arora, head of agent infrastructure at Vercel, told CNBC. “In its first full week after launch, the daily token volume grew by nearly 27x and the number of customers using it grew by nearly 80x.”

“Price is doing a job here,” Arora said. “When the job doesn’t require the best model, teams start moving it to something cheaper enough, and the latest wave of models coming out of China is winning that trade.”

China’s open-source models can be “60% to 90% cheaper” than Anthropic and OpenAI’s leading models, Justin Summerville, who works with data and analytics at OpenRouter, told CNBC.

OpenAI and Anthropic have been asked to comment on it.

While Claude and ChatGPT still dominate in terms of usage on LaunchLemonade, the AI ​​agent platform for managed industries, GLM 5.2 is now in the top five models on the platform, LaunchLemonade CEO and founder Cien Solon told CNBC.

“Chinese models like Z.ai and [Alibaba‘s] Qwen is becoming a choice for companies [as] they offer an attractive combination of performance and cost for certain workloads,” said Cien. “Enterprises with mature AI strategies are increasingly willing to use them where it makes technical or commercial sense.”

The performance of Chinese AI models is also increasing.

Although they are often “half the cost” of US rivals, they operate “closer to the top models on the American frontier,” Brookings’ Chan said, estimating that they are currently “six to nine months” behind top US rivals.

“The new open source models are working well and are proving to be able to do everyone without complex LLM jobs,” said Summerville.

The GLM 5.2 came within a percentage point of Anthropic’s Opus 4.8 in the closely watched single-agent benchmark, at about a fifth of the cost. Some researchers say that GLM 5.2 can perform on par with top US labs in some cyber benchmarks.

The switch to DeepSeek V4 has increased the performance of many of Lindy’s use cases, Crivello said in a post on X.

“We’re seeing companies increasingly motivated to turn to cheaper AI stacks that they can control and adapt, and given the nature of open models and the weight that often means using Chinese options,” Yacine Jernite, head of machine learning at Hugging Face, told CNBC.

“There is a real risk that users will be stuck when they have to choose between effective but expensive US proprietary models whose prices and accessibility can change quickly, or using Chinese models as the only viable option whenever they want to control costs or own their AI stack.”