Acute Myeloid Leukemia (AML) is a complex hematologic malignancy where precise subtype classification is critical for targeted treatment and improved patient outcomes. This study explores the potential of ConvNeXt, an advanced convolutional neural network architecture, for high-resolution peripheral blood smear image classification into AML subtypes. The dataset from a specialized hematopathology center provides a diverse and representative sample, addressing gaps in global leukemia diagnostics. A comprehensive deep learning pipeline was developed, integrating Stochastic Weight Averaging (SWA) for stability, Mixup data augmentation for enhanced generalization, and Grad-CAM for interpretability, ensuring biologically meaningful feature visualization. The ConvNeXt model achieved a state-of-the-art accuracy of 95%, surpassing traditional CNNs (ResNet50, 91%) and transformer-based models (Vision Transformers, 81%), demonstrating its superior feature extraction and classification capabilities. Grad-CAM visualizations provided biologically interpretable heatmaps, enhancing trust in computational predictions and bridging the gap between AI-driven diagnostics and clinical decision-making. Additionally, ablation studies highlighted the contributions of data augmentation, optimizer selection, and hyperparameter tuning, reinforcing the model’s robustness and adaptability. This study advances the role of AI in hematopathology by combining high classification performance, explainability, and scalability, paving the way for equitable and efficient AML diagnostics. Using clinically aligned evaluation metrics (accuracy, F1-score, and ROC-AUC) ensures its practical applicability, establishing a strong foundation for future AI-driven leukemia classification across diverse and underrepresented populations.