Phishing attacks have become a widely widespread cybersecurity threat, exploiting consumer vulnerabilities by mimicking legitimate web sites to thieve touchy facts. current detection structures regularly fail to adapt to evolving attack styles, in particular 0-day phishing attacks. This paper introduces AntiPhishStack 2.0, an enhanced phishing URL detection framework that leverages transformer-primarily based contextual analysis, GRUs with attention mechanisms for sequential dependencies, and superior characteristic engineering techniques. The machine employs a hybrid two-phase architecture and a CatBoost meta-classifier to obtain advanced detection accuracy, efficiency, and adaptability. Experimental effects on benchmark datasets show a detection accuracy of 98.01%, outperforming conventional models. The inclusion of light-weight deployment alternatives, along with TensorFlow Lite and ONNX, ensures actual-time applicability even in resource-limited environments. AntiPhishStack 2.0 sets a new benchmark in phishing detection, imparting sturdy defenses against sophisticated phishing strategies.