AnswersPublished19 sources
DeepSeek Narrows the U.S.–China AI Benchmark Gap to 3%
Bloomberg Intelligence put the U.S. lead over China in AI benchmark scores at about 3% after DeepSeek’s V4.1 Flash release, down from roughly 15% earlier in 2026 and 9% in May.
Published byEdited with GPT-6 LunaImages generated with GPT Image 2
Research answer
Key takeaways
- Bloomberg Intelligence put the U.S. lead over China in AI benchmark scores at about 3% after DeepSeek’s V4.1 Flash release, down from roughly 15% earlier in 2026 and 9% in May.
- V4.1 Flash ranked sixth on LiveBench with 81.1 points, compared with Anthropic’s 83.4; only three of the leaderboard’s top 15 models were Chinese.
- Better scores and hardware optimization could help Chinese developers compete, but low margin sales and price competition remain obstacles to profitability.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: According to Bloomberg Intelligence senior analyst Robert Lea’s October 5, 2026 report, how has the US–China AI model performance gap narrow. Article summary: Lea’s finding is a sharp convergence in *benchmark performance*, not proof that Chinese AI companies have caught up commercially: the reported US lead fell from about 15% earlier in 2026 to 9% in May and roughly 3% after. Topic tags: general, news, general web, user generated. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts w
Bloomberg Intelligence senior analyst Robert Lea reported that the U.S. lead over China in AI model benchmark scores had narrowed to about 3% after DeepSeek released V4.1 Flash in September. The estimate was roughly 9% in May and 15% earlier in 2026. It is a notable shift in benchmark performance, but not evidence that Chinese developers have matched U.S. companies commercially. 1
5
What DeepSeek’s LiveBench result shows
DeepSeek V4.1 Flash scored 81.1 on LiveBench and ranked sixth. Anthropic’s leading model scored 83.4, a difference of 2.3 points. Lea described DeepSeek’s performance as comparable to leading systems from Anthropic and OpenAI; the score comparison itself, however, is a result on one benchmark, not a verdict on every model capability or use case. 5
10
The result was reported as the highest LiveBench placement for a Chinese model since DeepSeek’s R1 drew global attention in 2025. But the leaderboard remained mixed: just three of its top 15 models were Chinese. Rankings can change, so one strong showing should not be read as broad Chinese dominance. 3
9
A narrower gap raises questions about chips and market share
Lea’s reported analysis links Chinese models’ progress to growing local expertise and tuning for domestic hardware. If those improvements continue, Chinese developers could become more competitive for users and market share even with restricted access to Nvidia chips. 1
4
That progress raises questions about how effective U.S. chip export restrictions will be over the long term. It does not, by itself, show that the restrictions have failed or establish how well domestic alternatives such as Huawei’s will perform. Benchmark scores measure model results, not the full effect of supply constraints or hardware strategies. 2
22
Why stronger benchmarks may not mean stronger profits
Model performance is only one part of a business. Bloomberg Intelligence’s outlook, as summarized by other reports, points to low-margin token sales and intense price competition as challenges for Chinese AI companies, with sustainable profitability potentially taking until around 2030. The precise timing is an analyst outlook, not a certainty. 6
9
The implication is that a smaller benchmark gap can strengthen the competitive case for Chinese models without resolving the economics of building and serving them. How much market share that performance translates into—and whether it can support lasting profits—remains a separate question.
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "DeepSeek Narrows the U.S.–China AI Benchmark Gap to 3%"?
Bloomberg Intelligence put the U.S. lead over China in AI benchmark scores at about 3% after DeepSeek’s V4.1 Flash release, down from roughly 15% earlier in 2026 and 9% in May.
What are the key points to validate first?
Bloomberg Intelligence put the U.S. lead over China in AI benchmark scores at about 3% after DeepSeek’s V4.1 Flash release, down from roughly 15% earlier in 2026 and 9% in May. V4.1 Flash ranked sixth on LiveBench with 81.1 points, compared with Anthropic’s 83.4; only three of the leaderboard’s top 15 models were Chinese.
What should I do next in practice?
Better scores and hardware optimization could help Chinese developers compete, but low margin sales and price competition remain obstacles to profitability.