To address the issues of high latency and excessive system overhead in Practical Byzantine Fault Tolerance (PBFT) for consortium blockchains, we propose an improved PBFT algorithm, GPBFT: a dynamic reputation and group-optimized consensus algorithm for large-scale consortium blockchains. The algorithm reduces the time complexity from O( \(N^2\) ) to O(N) by optimizing the consensus protocol and adopting a group-based broadcasting strategy to enhance its applicability in large-scale network environments. Subsequently, we design a reputation value calculation model based on propagation delay attenuation to dynamically update node reputations according to their behavior. Furthermore, we implement an enhanced Gaussian Mixture Clustering algorithm to dynamically adapt grouping strategies, thereby improving both system stability and the accuracy of leader node selection. Experimental results demonstrate that GPBFT reduces latency by approximately 80% and improves throughput by 2–4 times compared to PBFT, while effectively reducing communication costs and mitigating the impact of malicious nodes.