<p>Pivot-based machine translation (PBMT) provides a viable solution for language pairs with limited or no parallel data. However, traditional PBMT systems often rely on fixed pivot languages and models, limiting their flexibility and performance. To address these limitations, we propose two novel frameworks that harness the growing availability of open-source multilingual machine translation (MMT) models. Our first framework, fixed pivot and dynamic translation model (FP-DTM), employs a fixed pivot language while dynamically selecting the most appropriate translation model for the target language from a pool of pre-trained MMT models. The second framework, dynamic pivot and translation model (DP-DTM), extends this flexibility by dynamically selecting both the pivot language and the translation model based on the specific language pair and available MMT models. Through extensive experiments, we demonstrate the effectiveness of both frameworks in improving translation quality, especially for low-resource language pairs. Our approach offers a promising direction for future research in PBMT, enabling more robust and adaptable translation systems.</p>