This study adopts Hofstede’s cultural dimensions to test four large language models (ChatGPT-4, Gemini 1.5 Pro, Ernie 4.0, and Deepseek-R1) across Chinese and American cultural contexts with Chinese and English prompts. Through 16 experimental conditions and 16,000 samples, it analyzes the cultural tendencies of LLMs. Findings reveal that LLMs can reflect macro cultural differences between China and the U.S., but their representation of American cultural values remains more stable across languages, while representation of Chinese culture is more language-dependent and variable. Model outputs differ substantially from real human data, with higher alignment when simulating American culture. The study discusses the alignment between generative artificial intelligence and human language culture and its implications for multicultural ecology and intercultural understanding.