Omnidirectional images are widely used in various fields such as virtual reality (VR), augmented reality (AR), and panoramic photography. However, most existing reference-free (NR) omnidirectional image quality assessment methods provide a single quality metric and are unable to identify the types of distortion that may exist in omnidirectional. This prevents subsequent image restoration tasks from automatically selecting an appropriate restoration method based on the distortion type. Furthermore, these methods often require extensive training resources. To address these two issues, we propose an ordinal prompt that can assess the quality level of omnidirectional images and classify distortion types. To better extract the distortion information of images, we further propose a proportional viewport sampling method that adapts to human browsing patterns. We conduct extensive experiments on two mainstream datasets (CVIQD and OIQA) and compare our method with state-of-the-art methods in terms of correlation coefficient, accuracy, and generalization ability. Various experimental results show that when trained with few parameters, our method still outperforms existing methods and has better generalization ability to different datasets and distortion types. The models and code are available on GitHub at https://github.com/w-qhai/MTIQA360 .