Abstract: Accurate prediction of endpoint carbon content in converter steelmaking is a key technology for enhancing process control, reducing costs, improving efficiency, and advancing intelligent steelmaking. However, traditional prediction models often suffer from challenges such as difficulty in obtaining mechanism parameters and limited interpretability. To address these issues, this study proposes a hybrid model (HM) for predicting endpoint carbon content that integrates both mechanism-based modeling and data-driven methods. First, the parameters of the mechanism model (MM) are optimized using a genetic algorithm (GA) to improve its effective representation of the converter smelting process. Subsequently, a nonlinear coupling between the optimized MM and a random forest (RF) model is established to construct the hybrid mechanism–data fusion framework. Validation using actual production data from S Steel Plant demonstrates that the HM outperforms traditional mechanism-based and purely data-driven models. Within the error intervals of (−0.01%, 0.01%) and (−0.02%, 0.02%), the hit rates reached 84.4% and 98.8%, respectively. The mean absolute error (MAE) and root mean squared error (RMSE) were 5.70wt‰ and 7.27wt‰, respectively. The results indicate that the proposed HM achieves high prediction accuracy and robustness while maintaining strong mechanistic interpretability, offering a promising approach for endpoint control and intelligent optimization in converter steelmaking.