The authors examine whether large language models (LLMs) can extract contextualized representation of Chinese public news articles to predict stock returns. Based on representativeness and influences, the authors consider seven LLMs: BERT, RoBERTa, FinBERT, Baichuan, ChatGLM, InternLM, and their ensemble model. The authors show that news tones and return forecasts extracted by LLMs from Chinese news significantly predict future returns. The value-weighted long-minus-short portfolios yield annualized returns between 35% and 67%, depending on the model. Building on the return predictive power, the authors further investigate its implications for information efficiency. The authors show the assimilation speed of the LLM signals is two days, and they contain fundamental information. The signals can be especially helpful under higher frictions, when firms have less efficient information environments, more complex news, and higher retail holdings. Interestingly, heterogeneous investors load their future trades oppositely on LLM signals upon news releases. These findings suggest LLMs can be helpful in processing public news, and thus contribute to overall market efficiency.