Stanford AI Index report: China has nearly caught up with the US in AI model quality with lower investment

The US lead over China in the AI model race has nearly disappeared. A new Stanford AI Index report shows that the gap between the best American and Chinese model has narrowed to 2.7%, even as the US maintains a huge lead in private investment volume.

The Gap Has Nearly Disappeared

Back in May 2023, the difference between American and Chinese models on key benchmarks ranged from 17.5 to 31.6 percentage points.

Now it's no longer about double-digit gaps, but single-digit percentages. As of March 2026, the best US model leads the Chinese one by only 2.7%, and in February 2025, DeepSeek-R1 briefly matched the US leader.

This is an important shift: the debate is no longer whether China can catch up with American labs, but who will be first in any given release cycle. Competition at the top has also intensified. Stanford notes that American and Chinese models have switched places several times since the start of 2025.

In the current snapshot, Claude Opus 4.6 from Anthropic scored 1503 points in Arena, while Dola-Seed-2.0-Preview from ByteDance scored 1464.

The gap between them looks significant only on paper: at market level, this is already competition within one league, not a chasm between two ecosystems.

Where China Is Already Ahead

The model leaderboards themselves are only part of the picture. Stanford shows that China is growing stronger not only in model response quality, but also in metrics that build long-term technological momentum: publications, patents, citation impact, and industrial automation. These layers typically determine how quickly a country can replicate leaders' successes, scale development, and turn research results into products and production. So looking only at chatbot rankings is no longer enough.

• China accounts for approximately 69.7% of the world's AI patents.
• The country's share of global AI publications is 23.2%.
• The share of Chinese papers in the top-100 most cited AI articles grew from 33 in 2021 to 41 in 2024.
• In 2024, China accounted for 54% of all industrial robot installations worldwide.

This doesn't mean the US has lost its lead in every direction. According to Stanford, in 2025 American organizations released 59 notable AI models against 35 for China, and the US remains stronger in more significant patents and infrastructure. But the overall trend is different: China has stopped being a country that merely copies or rapidly catches up. It has built its own large-scale AI engine that systematically sustains frontier results.

Money No Longer Explains Everything

The most uncomfortable conclusion from the report for the US is that money no longer provides the advantage it once did. In 2025, private AI investment in the US reached $285.9 billion versus $12.4 billion in China. Formally, that's a 23-fold gap. By entrepreneurial activity, the difference is also large: the US saw 1953 newly funded AI companies, China saw 161. But in the quality of the best models, this financial gap is barely visible.

There is an important caveat. Stanford directly notes that private investment alone likely underestimates the real scale of Chinese spending: state-directed funding channels directed approximately $184 billion into the country's AI companies between 2000 and 2023. In other words, comparing $285.

9 billion against $12.4 billion doesn't mean China is building AI with almost no money. It means something different: visible venture capital statistics already poorly describe the real balance of power.

Meanwhile, US advantages haven't disappeared. The country still produces more top models, maintains leadership in weightier patents, and remains the primary node of AI infrastructure: it accounts for 5427 data centers, more than ten times higher than any other country. But even this foundation has stopped guaranteeing a comfortable technological lead.

An additional signal is the 89% drop in the flow of AI researchers and developers to the US since 2017.

What This Means

The main conclusion is simple: the global AI race is no longer just about who has more venture capital. China has shown it can approach the frontier through a combination of research, industrial base, and alternative financing mechanisms. For investors, corporate clients, and regulators, this changes the outlook: Chinese models can no longer be considered second-tier. Following OpenAI, Anthropic, or Google is no longer sufficient—Chinese labs and their ecosystem have become a mandatory part of the picture.