Leukemia is a fatal medical condition that should be diagnosed as early as possible in order to ensure proper treatment and better patients’ condition. Conventional diagnosis methods utilizing blood smear microscopy require much time for manual analysis and the inherent potential of error. This work explores the possibility of deep learning-based models for automated leukemia detection from blood smear images. A dataset of 2300 images, collected from hospitals and online sources, was used to train and test four convolutional neural network (CNN)-based models: Inception V3, ResNet, DenseNet, and MobileNet. The models were evaluated based on their ability to predict leukemia accurately, with Inception V3 having the highest accuracy at 96.7%, followed by ResNet at 93.2%, DenseNet at 90.34%, and MobileNet at 86.76%. AUC-ROC was also calculated for the models, where Inception V3 led with an AUC of 0.986, thus showing excellent discrimination between leukemia-positive and negative samples. MobileNet was associated with the fastest inference time of 0.28 s and the smallest model size of 16 MB. The results show that although Inception V3 is the most accurate, MobileNet presents a viable alternative for rapid, efficient deployment on mobile devices or embedded systems. These findings point to the need to select the appropriate model based on the application’s requirement for accuracy or computational efficiency, which has significant implications for enhancing early leukemia diagnosis and patient care in clinical and low-resource settings.