Deploying Deep Learning models in real-world requires considerations that are generally overlooked during training. In real-world scenarios, where the input data cannot be controlled, it is important for a model to identify when a sample does not belong to any known class. This issue is accomplished using out-of-distribution (OOD) detection, a technique designed to distinguish unk nown samples from those in the in-distribution classes. These methods mainly rely on model output scores or intermediate features to calculate OOD scores. However, the gradient space remains underexplored for this task due to the complexity involved in generating a suitable gradient representation. However, the gradient space is generally known for providing a richer representation of the model s knowledge or uncertainty regarding the expected output for the input sample, which is fully aligned with what is expected for an OOD detection task. Based on this idea, in this work we propose a new family of methods using gradient features, named GradVec, which uses the gradient vector as input representation for different OOD detection methods. The main idea is that the model gradient presents, in a more informative way, the knowledge that a sample belongs to a known class, being able to distinguish it from other unknown ones. GradVec methods do not change the model training procedure, with no additional data needed to adjust the OOD detector, and it is applica ble to any pre-trained model. We evaluated our model in different tasks, such as image classification, text classification, and semantic segmentation. Our ap proach presents superior results in different scenarios for OOD detection in image classification and sentiment analysis, reducing FPR95 by 88.17 percent and 56.91 percent, respectively.