The present research، aimed at developing a comprehensive flood modeling toolkit based on hybrid dual-AI frameworks، evaluated six models، including two individual models (FCMR and GRNN) and four hybrid dual models (FCMR-NARX، FCMR-ACOR، GRNN-NARX، and GRNN-ACOR) at five hydrometric stations in the Kraj watershed (Gachsar، Sira-kalvan، Sira-karaj، Neshatarud، and Morud) over a 30-year period from the beginning of the water year 1990 to the end of the water year 2018. The results، based on statistical criteria (R، NSE، RMSE، MAE)، indicated that hybrid models based on FCMR (particularly FCMR-NARX and FCMR-ACOR) have the highest accuracy. However، statistical tests did not show a significant difference in accuracy between these models and the individual GRNN model. On the other hand، hybridizing the GRNN model with NARX and ACOR catalysts due to structural incompatibilities reduced its accuracy. Considering the computational complexity، long training time، and higher implementation costs of hybrid models، the individual GRNN model is proposed as an optimized option with economic and operational justification for use in flood warning systems and water resource management in similar watersheds. This study emphasizes the necessity of coordination between the base model structure and the hybridization method to achieve effective performance improvement.