Sequential recommendation focuses on predicting the next item based on user historical behavior. Contrastive Learning (CL), with its capacity to optimize data representation and alleviate data sparsity issues, has been increasingly utilized in sequential recommendation. However, the majority of existing CL approaches rely on manual data augmentation techniques, such as item substitution and sequence reorder. These simple strategies, while intended to enhance model training, can inadvertently distort the underlying user intent, complicating the model’s ability to accurately capture and learn nuanced representations of user behavior. On top of that, these simple strategies might not work well for all kinds of datasets. In response to the aforementioned challenges, we propose a novel framework for sequential recommendation. While using data augmentation, our work also introduces adaptive model augmentation to adapt to various datasets. This innovative step allows for seamless adaptation to diverse datasets, ensuring that the framework can effectively harness the benefits of both data and model augmentation to accurately capture user behavior representations. Furthermore, we introduce a discriminator with gating mechanism to distinguish between data augmented sequences and their original versions. This feature is crucial in maintaining the model’s sensitivity to significant augmentations, thereby preventing many potential biases that might mislead the model’s understanding of user behavior. Comprehensive experiments on three public datasets demonstrate the effectiveness of our proposed framework compared to the state-of-the-art sequential recommendation models.