This paper proposes machine learning-based option pricing models that incorporate firm characteristics. We employ two semi-parametric models, one that uses machine learning to predict the implied volatility and the other that corrects the pricing error of a parametric model, and explore a set of 111 firm characteristics for enhancing the models’ pricing performance. Our empirical analysis is conducted using big data featuring 15,247,956 stock option observations in the period January 1996 to December 2021. We find that both semi-parametric models outperform the parametric one even without firm characteristics, whilst firm characteristics significantly enhance the performance of these models. Conditional skewness, Dimson’s Beta, dividend yield, annual stock return, and downside beta are found to be the most important firm characteristics.