China is having another AI moment

The US’ lead over China in artificial intelligence may be at its smallest in over a year. When China disrupted the AI race in January 2025 with the release of DeepSeek R1, it erased US$1 trillion (S$1.29 trillion) from US capital markets. Nvidia, a chip firm, briefly shed 17 per cent of its value; the NASDAQ sank by 3.1 per cent in a day.

US investors were troubled not only because Chinese AI was good but because it was being given away for free. The uproar soon faded. Since then, market valuations around the world have hinged ever more on the promise that AI will be both revolutionary and profitable.

Now Chinese labs are unsettling their US rivals anew in the race to monopolise the market for models. On June 13, a Beijing-based lab called Zhipu, or Z.ai, announced its latest system, GLM 5.2, promising “a step closer to frontier intelligence for everyone”. It is the most capable Chinese-trained model to date and runs at less than a tenth of the price of Anthropic’s latest release, Fable 5. And as with other Chinese models, the weights, or parameters, that enable GLM 5.2 to function have been publicly released.

In recent weeks, US companies have been grappling with soaring AI costs, sometimes ranging into the thousands of dollars per employee. Some firms are setting budgets for tokens (bits of text processed by a model). Then on June 12, the Trump administration banned non-Americans from using Fable 5, leading Anthropic to switch off access for everyone.

For the first time, access to frontier AI rests on the US government’s say-so. All this may give users reasons to look at alternatives to US AI. Many will find GLM 5.2 capable and affordable, and welcome that it is out of the Trump administration’s reach.

Start with capability. Artificial Analysis, a research firm, ranks GLM 5.2 as the most intelligent open-source model on the market. GLM 5.2 takes an impressive fourth place on its overall list, behind OpenAI’s ChatGPT 5.5 and ahead of Google’s Gemini bot. The model has surprised everyone.

Earlier this year, Chinese developers were pessimistic about the prospect of their models outclassing US ones before 2030. After Zhipu’s release, Elon Musk, a very rich man, wrote on X, his social media site, that he expects China to match the abilities of the current frontier by early next year. It “won’t take that long”, Tang Jie, Zhipu’s co-founder, shot back.

Unlike in the DeepSeek moment, US markets have so far shown little interest in GLM 5.2. This is partly because it has become more difficult to accurately assess the ability of Chinese models. To arrive at its estimates, Artificial Analysis scored GLM 5.2 on dozens of benchmark tests, which use exam-like questions to evaluate a model’s smarts.

The US, via Anthropic, keeps its edge in performance. Fable 5 is about 17 per cent cleverer than GLM 5.2 across an average of benchmark tasks. The other important metric is how long it took GLM 5.2 to reach this level of intelligence. A comparable Western model to GLM 5.2 was released in February, or about four months ago.

In reality, the US lead is probably bigger than four months. Open-source models, many of them Chinese, tend to score better on public benchmarks than private ones, says Havard Tveit Ihle of the Norwegian Defence Research Establishment, a think-tank in Norway. The questions used in public benchmark tests are published, whereas those who apply private benchmarks keep their evaluations secret.

Analysis by Tveit Ihle published before GLM 5.2 found that Chinese models were about four to six months behind US ones on public tests. But on private tests, the US’ lead nearly doubled, to eight to 10 months. A study by the US government, released in May, identified a similar gap. Tveit Ihle says Chinese labs appear, possibly unwittingly, to “teach to the test”.

On two private benchmarks tested so far, GLM 5.2 shows the same hallmarks: It is about seven months behind on WeirdML, a measure of unusual machine-learning tasks that need careful reasoning to solve, and fully a year behind on SimpleBench, which evaluates common sense by trying to trick models.

The pattern is not consistent, however. A new exam released by Artificial Analysis on June 19 tests models on office-worker tasks, like sifting through messy files and evaluating conflicting information. GLM 5.2 could not have trained for the evaluation. Yet it outperformed ChatGPT 5.5, which is just two months old. These results suggest that America’s lead remains steady, says Tveit Ihle, but are also evidence the gap is not widening as some had expected it would.

What is especially surprising about GLM 5.2 is that it succeeds in tasks that tend to trip up its peers. Chinese models often excel in fields with clear right or wrong answers, like maths and coding. But they tend to fall down on problems that are open-ended or that require sustained independent judgment.

That pattern reflects one of the largest challenges facing researchers in China. Export controls on advanced chips have left Chinese labs short of the computing power needed to train the strongest models. So they tend to make up ground in post-training: fine-tuning models to behave in particular ways or solve certain kinds of problems, including on data allegedly harvested from US systems through a process called “distillation”.

Given the uncertainties surrounding the true capabilities of Chinese models, next, consider whether they are actually cheaper than their US rivals. DeepSeek charges just 87 US cents per one million output tokens for its v4 model, whereas Anthropic charges US$50 for the same on Fable 5. Such prices might have a growing appeal in the US, where token costs at some firms have run out of control. In June, DeepSeek saw a sharp rise in US firms paying for its services, according to Ramp, an invoicing company. Microsoft is reportedly considering using the Chinese lab’s model in its flagship Copilot chatbot. Yet this most important assumption, that Chinese AI is cheaper, can frequently be wrong.

Though Chinese models are becoming more capable, they are generally not becoming more efficient. Chinese models use many more tokens to think through their answers. A study updated this month by Du Zheng of Georgia Tech and co-authors shows that given the same tasks, a DeepSeek model used 23 times more tokens than its OpenAI rival to achieve basically the same result. Because of these large differences in efficiency, the correct way to compare models is not price per token but the total cost of all the tokens used. Using this metric, on a benchmark designed to test software engineering, GLM 5.2 ended up costing more than competing systems from Anthropic and OpenAI.

In addition to capability and cost, a third selling point is now top of mind for AI users: reliability. Zhipu released its model at 5.21pm Beijing time on June 13, one day after the Trump administration told Anthropic that it was banning non-Americans from using Fable 5.

“Our attitude is one of radical openness,” Tang declared. He also blasted “external blockades”, such as the one imposed by Anthropic and the US government, saying they made AI systems “subject to revocation at any moment”.

Most Chinese models are released open-source, meaning they can be downloaded and run on local hardware, out of reach of governments or the labs themselves. The US government could one day impose limits on the domestic use of Chinese AI. Two congressional committees are currently investigating US tech firms for using Chinese models. China’s labs face other limitations to their reliability: A shortage of computing power means they often run into service interruptions or slow down in periods of high traffic.

As the AI race speeds up, however, regulators everywhere will be faced with new challenges to safety and security. The risk of sudden government intervention may grow on both sides of the Pacific Ocean. Fable 5 was powerful enough to prompt such a response from the White House. That Chinese models are not, for now, subject to similar regulatory risk suggests China’s government is not yet alarmed enough to act. That may be some of the clearest evidence that they remain behind their rivals. © 2026 THE ECONOMIST NEWSPAPER LIMITED. ALL RIGHTS RESERVED.