I employ Large Language Models (LLMs), including BERT and an OpenAI model, to extract information from news articles and predict option returns. LLM-based news portfolios achieve annualized Sharpe ratios of up to 3.15 and outperform news portfolios constructed using other methods. Commonly used observable and latent factors in the stock and options markets do not explain the returns of these news portfolios. Firm-specific and pharmaceutical-related news play key roles in predicting option returns. These portfolios perform better for firms with high R&D expenditures or high stock volatility.