Schizophrenia (SZ), an intricate mental health condition, is defined by a variety of symptoms, such as delusions, disorganized thinking, hallucinations and cognitive impairments. Its heterogeneous nature, which varies significantly across individuals in terms of symptom severity, progression, and response to treatment, makes diagnosis particularly challenging. The electroencephalogram (EEG), which records brain electrical activity through scalp electrodes, plays a vital role in SZ research. Its capacity to detect nuanced variations in neural activity using high temporal resolution offers valuable insights into brain function. Various machine learning (ML) and deep learning (DL) models have been designed to improve the accuracy of SZ detection, aiming for early diagnosis and better patient outcomes. This research introduces a hybrid DL model (SZ-Net), combining Convolutional Neural Networks (CNN), Bidirectional Long Short-Term Memory (BiLSTM), and Bidirectional Gated Recurrent Units (BiGRU) with attention mechanisms to analyze EEG signals for SZ detection. The model is evaluated on two benchmark datasets—the EEG dataset obtained from M.V. Lomonosov Moscow State University (MSU), comprising 39 healthy and 45 SZ patients, and the Kaggle EEG dataset from a basic sensory task in SZ, comprising 32 healthy and 49 SZ patients. Our proposed framework achieves accuracy rates of 98.21% and 99.27%, respectively. It outperforms pioneering results, showing its exceptional capability in EEG-based SZ diagnosis. These results highlight the framework’s capability as a dependable solution for early diagnosis and intervention. The hybrid architecture effectively captures both spatial and temporal patterns in EEG signals, enabling a more comprehensive analysis. Additionally, the study opens avenues for future research focused on enhancing model efficiency, real-time implementation, and interpretability, with the goal of providing accessible and accurate diagnostic solutions for SZ.