<p>Federated learning is a powerful machine learning framework that enables collaborative model training while preserving data privacy. Blockchain-based federated learning has emerged as a decentralized solution to enhance system verifiability and security. However, existing blockchain-based federated learning approaches suffer from limited fault tolerance and high communication overhead, posing significant challenges for large-scale deployment. To address these issues, we propose a quantum-enhanced blockchain federated learning framework. It integrates quantum Byzantine agreement, enabling consensus even when nearly 50% of clients are malicious, significantly improving fault tolerance. Additionally, we leverage matrix product operators for model compression, reducing communication overhead by up to 90% while maintaining model accuracy. We design a Byzantine-resilient aggregation algorithm that effectively mitigates adversarial attacks and enhances privacy protection. Experimental results on two benchmark datasets show that even with 40% of clients being malicious, the proposed method maintains strong performance, significantly surpassing traditional approaches in terms of effectiveness, robustness, and security.</p>