The gap between American and Chinese AI is closing fast. Chinese models narrowed the performance shortfall to a record-low 6% in June, down from 9% in May, according to Bloomberg Intelligence. The research arm said the shift raises real questions about how long US supremacy in AI will hold.
The lead used to look comfortable. Over the previous 12 months, the gap averaged 10% to 15%, BI said. Then Zhipu AI’s GLM-5.2 reached the top of the global ranking in agentic coding, and the maths changed.
It was not a solo act. Chinese models took two slots in LiveBench’s global ranking of large language model performance last month. The arrival of Moonshot’s Kimi “highlights that Zhipu’s success was not a one-off,” said Robert Lea, a Bloomberg Intelligence analyst, in a report.
Two philosophies, one race #
The two sides are not playing the same game. American labs lean on plentiful Nvidia chips to push performance as far as it will go, BI said, and pay a heavy compute bill for it. Chinese firms take a leaner, more efficient path. They may trade away some performance to do so.
Which wins is still an open question. “Time will tell which approach yields greater results,” the analysts wrote. For now, the efficient approach is closing the distance quickly.
Why the number lands hard #
The timing gives the figure weight. A cheap Chinese model has spent weeks unsettling US investors and officials. Kimi K3’s launch set off a sharp market reaction, echoing DeepSeek a year earlier. US officials have since floated sanctions over alleged model theft.
A single percentage point will not settle a technology race. But the direction is clear. The story has moved from whether China can keep pace to how small the gap can get.
Get the TNW newsletter #
Get the most important tech news in your inbox each week.