Three percent. That’s all that’s left of America’s lead over China in artificial intelligence.
Top Chinese AI models now trail their US rivals by just 3% on benchmark scores, according to a Bloomberg Intelligence report published Monday, October 5. The gap sat at roughly 9% in May and around 15% earlier this year, and the move followed the September release of DeepSeek’s V4.1 Flash model. Senior analyst Robert Lea, who wrote the report, said the trend points to further market-share gains for Chinese labs.
The 3% figure is a record low, and it lands in a year when Washington has treated AI supremacy as a national project.
How the US China AI lead collapsed this fast
DeepSeek’s V4.1 Flash ranked sixth globally in September on LiveBench, making it the highest-ranked Chinese model since the startup’s reasoning model R1 broke through in 2025. LiveBench grades models on their responses to questions, puzzles, and analysis tasks, roughly the way an IQ test grades a person.
China’s ascent, according to the report, comes from two things: deepening AI expertise and researchers’ growing ability to optimize their models for domestic hardware. That’s the part Washington won’t like. US export restrictions on technology like Nvidia chips were designed to curtail Chinese AI advances and stop Huawei from building viable alternatives. Lea’s report casts doubt on whether that strategy still works.
In his words, the progress “casts further doubt on the long-term sustainability of US technological supremacy in AI.”
The two countries are fighting over a technology that drives economic productivity, military strength, and global influence. Chinese labs are doing it cheap. Their lower-cost models are closing the gap not just on benchmarks but on user numbers, at the same time Anthropic and OpenAI are chasing trillion-dollar valuations in planned stock-market debuts by selling the promise that their capabilities are superior.
DeepSeek has done this before
This isn’t the first time DeepSeek has rattled Washington. The startup’s R1 release in January 2025 helped trigger a record rout in Nvidia shares, as investors suddenly questioned why US hyperscalers needed to spend hundreds of billions of dollars on chips if a Chinese lab could build something comparable at a fraction of the cost, Startup Fortune noted in its coverage. Nvidia erased $589 billion of market value on January 27, 2025, the largest one-day wipeout in US stock-market history. Its shares fell nearly 17% that day.
The stock bounced back quickly, and Bloomberg Opinion wrote at the time that the DeepSeek panic looked more like an aberration in an otherwise unbroken AI boom than a genuine Sputnik moment. The 3% benchmark gap suggests the aberration is turning into a pattern.
There’s also company around DeepSeek this time. The Huawei news keeps piling up alongside it: the Chinese giant struck a broad multiyear patent deal with Qualcomm on Monday covering 5G, AI, and computing technology, and its expected patent-licensing contract value has now topped $6.9 billion. Sanctions haven’t stopped it from monetizing the one thing it kept investing in — research.
What the 3% gap means for Washington
Here’s the uncomfortable part. The US government is reorganizing itself around the premise that America is comfortably ahead. President Trump announced a “Super Intelligence Force” on Sunday to coordinate federal AI efforts, led by Director of National Intelligence Jay Clayton and reporting directly to the White House. The whole bet is that the US stays in front.
But a 3% benchmark gap doesn’t read like a lead that a task force can defend. It reads like a market reality the White House is reacting to late.
The export-control argument is the one taking the most damage. If Chinese labs are closing a 15-point benchmark gap to 3 in under a year while largely locked out of the latest Nvidia hardware, the restrictions look less like a wall and more like a tax — one China’s best labs have learned to pay. Nobody in Washington will say that out loud. The Bloomberg Intelligence numbers are starting to say it for them.
And then there’s the pricing question. If Chinese models are 97% as good on benchmarks for a fraction of the cost, the argument for paying a premium for American frontier models gets thin. That’s the argument Anthropic and OpenAI are selling to IPO investors at trillion-dollar valuations. It’s a harder pitch when the gap is a rounding error.
Why this matters
The AI race was supposed to be America’s to lose. Nobody said it couldn’t be lost.
For the average American user, this shows up as cheaper, faster AI tools — Chinese models like DeepSeek’s remain free to use while US labs charge premium API rates and subscription prices. For the industry, it shows up as a valuation question: Silicon Valley’s trillion-dollar bet assumes capability commands a premium, and that assumption is bleeding out on the benchmarks.
For Washington, it shows up as a policy reckoning. The export controls, the task forces, the rebrand from “artificial intelligence” to “super intelligence” — all of it is built on the premise of a durable lead. A 3% gap is not a durable lead. It’s a rounding error with a flag on it.
Frequently asked questions
How far ahead is the US in AI compared to China?
Bloomberg Intelligence estimates the gap is now just 3% on benchmark scores, a record low, down from about 9% in May 2026 and 15% earlier in the year. The figure comes from a report by senior analyst Robert Lea published October 5, 2026.
What is DeepSeek’s V4.1 Flash?
DeepSeek’s V4.1 Flash is a fast, low-cost reasoning model released in September 2026. It ranked sixth globally on LiveBench that month, the highest placement for a Chinese model since DeepSeek’s R1 broke through in 2025 and triggered a record selloff in Nvidia shares.
Does this mean US chip export controls have failed?
Not entirely, but Bloomberg Intelligence says the gains raise questions about their long-term usefulness. China’s progress came from deeper AI expertise and optimizing models for domestic hardware rather than imported chips, which is exactly the outcome the restrictions were meant to prevent.
Why does a 3% AI gap matter?
Because the premium price tag on American frontier models gets harder to justify when Chinese rivals are nearly as capable for far less. It also complicates the case for Anthropic and OpenAI as they pursue trillion-dollar IPO valuations built on the claim of superior capabilities.
Sources: Bloomberg Intelligence (reported by Bloomberg Law, The Business Times, and The Edge Singapore), Startup Fortune, Reuters