Early identification of tomato leaf diseases is essential because of its major influence on crop yield. With the advances of computer vision and deep learning, detection solutions have been studied recently, benchmarking on some well-known datasets, such as PlantVillage. However, the coverage of one dataset alone is not sufficient to ensure the classification quality and generalizability of trained models. Hence, this work presents Tomato Leaf Disease of size 46070 (TLD46k), a benchmark dataset that combines PlantVillage, CCMT, and PlantDoc. We cross-validate the performance of YOLO11 models on those datasets. Eperimental results highlight the enhancement of the model’s performance when trained on TLD46k, i.e. up to 80.7% in accuracy and 81% in Macro- \({F}_{1}\) .