<p>High-performance concrete (HPC) is a low-carbon construction material that aligns with global sustainability goals focusing on climate change mitigation and reducing the carbon footprint of the construction industry. Because this industry contributes significantly to global CO₂ emissions, developing sustainable alternative materials that balance environmental, economic, and performance requirements is critical to advancing green construction practices. However, existing studies lack comprehensive prediction models that effectively balance key decision-making factors, including cost, strength, and carbon emissions. To address this gap, this study develops an evolutionary deep learning model, ASOS-NN-BiGRU, which integrates Neural Networks (NN) and Bidirectional Gated Recurrent Units (BiGRU) to process independent and sequential data in HPC mixtures. The model is optimized using the Auto-tuning Symbiotic Organisms Search (ASOS) algorithm to enhance compressive strength prediction accuracy. The developed model is further deployed to optimize HPC mixture designs under three key scenarios: minimizing overall carbon emissions, identifying the most cost-effective mixture without carbon fees, and determining the most cost-effective mixture considering potential carbon fees. Additionally, the Multi-Objective Auto-tuning Symbiotic Organisms Search (MOASOS) algorithm is employed to identify optimal low-carbon HPC mixtures. By integrating carbon pricing mechanisms and multi-objective optimization, this research provides a practical framework for sustainable concrete production that supports both the transition of the construction industry toward low-carbon materials and the development and implementation of effective carbon taxation policies. Experimental results confirm the model’s robustness and reliability, enabling decision-makers to design HPC mixtures tailored to specific sustainability and cost preferences while ensuring structural performance.</p>