RNAret is a lightweight and efficient pretrained RNA language model based on the Retentive Network (RetNet) architecture, which enables parallel training, low-cost inference, and strong performance for modeling long RNA sequences. It can be fine-tuned for various downstream tasks, making it accessible to academic teams with limited computational resources (with approximately 12 million parameters). Through evaluations across multiple sub-tasks related to RNA structure, function, and type, RNAret demonstrates its ability to capture the features of RNA sequences. This highlights the potential of leveraging advancements in large language models for biological applications, showcasing the efficacy of encoder-only architectures based on RetNet for addressing complex biological challenges.