Comparing U.S. vs. Chinese AI Model Performance 🇺🇸🇨🇳

What We're Showing

The performance of the top U.S. and Chinese AI models on LMSYS's Chatbot Arena from January 2024 to February 2024.

Chatbot Arena is an open platform for benchmarking large language models based on crowd-sourced user preferences.

Data comes from LYMSYS via Stanford University's 2025 AI Index Report.

China Is Narrowing The AI Performance Gap

Historically, the top U.S. models have consistently outperformed those from China. But the gap is closing fast.

In January 2024, the performance difference between the top U.S. and Chinese models was 103 points. By February 2025, that margin had shrunk to just 23 points.

The rapid catch-up is largely credited to the launch of Deepseek R1, an open-source Chinese model that delivered strong results reportedly using just a fraction of the compute resources of U.S. models, shaking confidence in U.S. AI leadership and causing stock market turbulence.