Six of the ten most popular AI models on a closely watched global leaderboard were built in China. That number, from a report covering events in mid-June, lands less than a month after the U.S. government ordered Anthropic to cut access to its two most powerful systems, Fable and Mythos. One of those six Chinese models, GLM-5.2 from the start-up Z.ai, hit the market days after the government demand.
It is nearly as capable as the restricted Anthropic models, but costs roughly an eighth as much for certain tasks. The price difference is the sharp edge of a broader shift.
Silicon Valley executives are watching the Trump administration lean toward regulating the technology. At the same time, U.S. businesses are scrambling to cut how much they spend on AI. The cheap Chinese models offer a way out.
Rehaan Ahmad, a co-founder of the Silicon Valley start-up alphaXiv, has used GLM-5.2 for more than a week. “With Fable restricted, the gap between the US and China is very slim,” he said.
The stakes are concrete. For about 18 months, U.S. companies have poured money into models from OpenAI, Anthropic, and Google. Those bills are now under pressure.
A cheaper alternative that performs nearly as well on standard tasks is not a curiosity. It is a business decision waiting to be made. Z.ai did not invent a new paradigm.
It built a model that lands on the same leaderboard and undercuts the competition on price. Two hurdles block wide adoption in the U.S. One is concern over the Chinese models’ ties to their government.
The other is a complaint that Chinese companies have used U.S. technology unfairly to build these cheaper systems. Neither barrier has stopped the models from arriving or from climbing the rankings. Six of the top ten are Chinese.
That is not a forecast. It is the current state of the leaderboard.
The timing matters. Anthropic shut down Fable and Mythos after an unexpected government demand. The company did not choose to pull them.
The U.S. government did. That left a gap in the market. Z.ai filled it within days.
The new model, GLM-5.2, did not need permission to launch. No one in the United States put restrictions on it.
The price advantage is winning converts among U.S. businesses that need to rein in spending. The report notes that Z.ai’s technology is winning those converts despite the concerns. Cost is a powerful argument.
When a Chinese model costs an eighth as much as a comparable U.S. model for the same job, the math does not care about politics. The wave is broader than one company.
Z.ai is on the cutting edge, but six Chinese models now sit on the leaderboard. The lock that OpenAI, Anthropic, and Google once held on the industry is not a lock anymore. The gap between the U.S. and China in AI capability is very slim, as Ahmad put it.
The gap in cost is already reversed. What is at risk is not just market share. It is the assumption that the most advanced AI will be built in the United States and sold at U.S. prices.
That assumption is breaking. A Chinese start-up released a nearly as powerful model for a fraction of the cost, and no one stopped it.
The U.S. government restricted its own companies’ models instead. The result is a leaderboard that now tilts toward China. Businesses in the U.S. have to decide.
Pay more for a restricted U.S. model, or pay less for a Chinese model that works and faces no restrictions. The report makes clear that some are already choosing the latter.
Sources
- Western open-weight models still can't beat China – Rest of World
- AI Leaderboard 2026: Rankings for 300+ Top AI Models by…
