Abstract: Electric arc furnaces (EAFs) have become a pivotal pathway toward sustainable steel production; however, challenges remain in optimizing the EAF process. This study presents a comprehensive framework integrating hybrid modeling with multi-objective optimization to simultaneously minimize cost (including carbon tax), electricity consumption, and smelting time. A novel hybrid modeling approach systematically combines metallurgical mechanisms, empirical knowledge, and machine learning algorithms (CatBoost, LightGBM, TabNet, TabPFN, SVR, DNN), with hyperparameters optimized via metaheuristic algorithms, including Teaching Learning-based Optimization, Firefly Algorithm, Whale Optimization Algorithm, and Fick's Law Algorithm. The optimized hybrid models demonstrated high predictive accuracy with R² values of 0.9523, 0.9318, 0.9092, 0.9321, and 0.8888 for endpoint carbon content, phosphorus content, temperature, electricity consumption, and smelting time, respectively. Industrial validation with 300 heats confirmed robust generalization capabilities. Furthermore, a systematic evaluation of seven state-of-the-art multi-objective optimization algorithms identified the Constraint-handling Two-Archive Evolutionary Algorithm (C-TAEA) as superior across five metrics (GD:0.0286, IGD:0.0432, GD+:0.0080, IGD+:0.0150, HV: 0.8114). The optimized solution yielded well-distributed Pareto-optimal alternatives, with the compromise solution achieving remarkable improvements over baseline operations: 206.12 CNY/ton cost reduction, 47.68 kWh/ton energy savings, 14.19 minutes time reduction, and 7.57% decrease in carbon emissions. This framework provides both methodological innovations in industrial process modeling and actionable insights for sustainable steel production, demonstrating significant potential for practical implementation in achieving economic, operational, and environmental objectives simultaneously.