The Multiple Choice Questions Generation (MCQG) method, which is contingent upon two pivotal sub tasks: Question Generation (QG) and Distractor Generation (DG), is widely regarded as one of the most effective and prevalent techniques for evaluating students’ learning outcomes. However, manually crafting such tests is time-consuming and laborious. Meanwhile, few scholars have examined the QG and DG tasks of MCQG together, and researchers rarely focus on utilizing a unified framework to evaluate MCQG output quality, particularly for Chinese reading comprehension multiple choice questions. Furthermore, the critical challenge in MCQG remains how to effectively use Pre-trained Language Models (PLMs) to balance the generation of correct content (such as question) and incorrect content (such as distractor). In order to address these challenges, we introduce a novel framework based on PLMs for MCQG. In this study, we explored our MCQG framework through three distinct fine-tuning approaches: Multi-model MCQG, Multi-task MCQG, and End2end MCQG. The experimental findings indicate that the End2end MCQG approach surpasses other methodologies, achieving a significant advancement with an average increase of 28.51 points in automatic evaluation metrics.