Chinese AI models make their way to the US and gain ground

Hong Kong – China is developing a hot new product in the US: artificial intelligence.

Raffi Krikorian, chief technology officer at San Francisco-based Mozilla, the company behind the Firefox browser, switched much of his day-to-day work to Chinese AI startup Mozilla’s Kimi K3 within days of the new powerful model being announced over a week ago.

He talked about K3, comparing it to Claude Fable, a much-praised, more expensive chatbot from San Francisco company Anthropic. Previously, he used another high-performance Chinese model, Z.ai’s GLM-5.2, for everyday tasks such as managing calendars, documents, and email.

Krikorian is one of a growing number of Americans adopting Chinese-made AI systems. Chinese-made AI systems are becoming more popular around the world because they are more affordable and more efficient. U.S. companies such as cryptocurrency exchange Coinbase have said they are switching to Chinese AI models to cut costs.

Its popularity has irritated some US tech giants, but unless it is banned outright, it will likely continue to appeal to independent software developers in the US and elsewhere.

U.S.-led regulations have prevented China from accessing the world’s most advanced technologies, including cutting-edge AI chips, and Treasury Secretary Scott Bessent has warned that further sanctions could be imposed to protect U.S. intellectual property.

US President Donald Trump’s administration on Wednesday accused Moonshot of using “secret” but not necessarily illegal methods to build K3 based on the Anthropic fable. Some U.S. politicians and AI companies such as Anthropic have accused Chinese startups of illegally “distilling” models to extract technology, but the Chinese government rejects this claim as “baseless.”

China’s AI models are cheap and ‘good enough’

The rapid rise of cutting-edge Chinese AI models this year signals further progress in the AI ​​race between China and the United States since early last year, when Chinese startup DeepSeek shook up the U.S. technology industry with an AI model that was as powerful as American AI models, but much cheaper, putting China on the map.

Experts say the latest AI models released in June by startups Z.ai (Zhipu) and Moonshot in July are almost as intelligent as Frontier models from OpenAI and Anthropic. Also in July, China’s Alibaba previewed the Qwen3.8 Max AI model. In April, DeepSeek released a preview of its latest V4 model, taking on OpenAI’s GPT and Google’s Gemini.

This constant progress is paying off. Kurt Meinhold, a technology executive based in Greensboro, North Carolina, said he increasingly prefers to use DeepSeek for tasks such as finding ways to generate leads and sales for his business.

“At the end of the day, most of us, more than 90 percent, don’t need myths and fables,” said Meinhold, who founded the digital legacy platform LilyList. “It’s like, I don’t need that, I need something that’s good enough.”

U.S. investment bank Goldman Sachs said in a research report in July that China’s AI models have reached a “critical stage” for widespread adoption, especially with the rapid rise in the use of “agent” AI globally, which requires AI models to autonomously perform multi-step, complex tasks, and increasing demand for cost-effective AI models.

Analysts including Alex Colville of the Australian Strategic Policy Institute say the use of AI “agents” dramatically widens the cost differential, as the price of AI is calculated per million tokens for both inputs and outputs.

“I think the Chinese model is pretty close in code and research,” Meinhold said. “If you can pay a few cents per token for a million outputs instead of $30, $40, $50, that’s good enough.”

Moonshot’s Kim K3 takes the lead, but China’s AI model has limitations

Based on data from the past month, the top five most popular models on OpenRouter, a platform that tracks data across AI models, were made in China.

In the week following K3’s July release, Kimi received more than 930,000 downloads, a 200% increase from the previous week, according to estimates from market intelligence firm Sensor Tower. It was downloaded approximately 86,000 times in the US, an increase of 387%.

K3 was so popular that Moonshot had to temporarily suspend new subscriptions as overwhelming demand neared capacity.

Still, while the Chinese model is becoming a serious competitor to the U.S., it lags behind the U.S. AI leader in overall full-range capabilities, said Anastasios Angelopoulos, co-founder and CEO of Arena, a platform that evaluates AI systems.

U.S. AI companies are also exploring further options to lower prices, including cheaper alternatives to competing models, added Yasir Attaran of the Center for Strategic and International Studies.

US policy could boost China’s AI model

While Frontier models from US companies like Anthropic and OpenAI are closed source, most Chinese AI models are open source. That means anyone can test and build.

Lian Jie Su of technology research and advisory group Omdia said Chinese model vendors are expected to leverage open source software to promote global usage and adoption.

Mozilla’s Krikorian said China’s open source model is becoming nearly as good as closed systems made by major American companies, saying, “The open frontier is becoming more and more made in China.” While the US considers restricting Chinese AI models, a group of US tech companies including Microsoft, Meta and Nvidia signed an open letter supporting “open” AI models published on Friday.

Other U.S. technology policies also sometimes give China an advantage, some experts say. China’s Z.ai released its GLM-5.2 model in mid-June, shortly after the Trump administration imposed export restrictions on Anthropic’s Fable and Mythos models, which took them offline for more than two weeks.

“Restricting the U.S. model could quickly open up an opening for Chinese competitors,” Arena’s Angelopoulos said.

China’s AI models aim for global expansion

In China, people and businesses are rapidly embracing AI technologies that have flourished with state support. Chinese tech companies such as Huawei and Tencent are also incorporating AI into devices such as smartphones, AI glasses, and humanoid robots.

Now, China is increasingly trying to dominate AI abroad as well. At a major technology summit in Shanghai, Chinese President Xi Jinping championed open source AI models and the importance of promoting greater global equity, and pledged China’s involvement in improving AI capabilities, especially in developing countries.

As with many other industries in China, intense domestic competition is driving companies to expand globally. China’s leading AI startups are currently raising even more capital to support their efforts, including through public offerings.

Chelsea Tam of investment research firm Morningstar said both China and the United States want to “encourage widespread adoption of AI ecosystems while protecting technologies that could significantly strengthen strategic rivals.”

“The competition is no longer just America vs. China. Chinese labs are also putting a lot of pressure on each other,” Angelopoulos said.

However, as in the US, heavy investments in AI have also raised concerns about the sustainability of these Chinese startups. Z.ai, for example, reported a 132% increase in revenue to 724 million yuan ($107 million) last year, but a 60% increase in net loss to 4.7 billion yuan ($694 million).

For now, China’s AI models are likely to continue to advance, including in the U.S., and “I highly recommend that anyone running a serious AI workload at least evaluate them,” Krikorian said.

___

O’Brien reported from Providence, Rhode Island. AP Business Writer Kelvin Chan in Toronto contributed to this report.