To　meet　the　service　requirements　for　delay-sensitive　tasks　and　improve　content　delivery,　in　this　paper,　a　deep　reinforcement　learning　(DRL)-based　joint　offloading　scheme　of　computing　and　content　tasks　in　cloud-edge-end　cooperation　networks　is　proposed.　We　use　a　minimum　delay　model　to　describe　the　task　offloading　problem,　where　the　same　requests　can　be　aggregated　to　lighten　the　server　load　and　the　in-network　caching　is　considered.　We　design　a　new　DRL　algorithm　to　achieve　intelligent　task　offloading　by　making　cooperative　caching　and　routing　decisions.　The　simulation　results　show　that　the　proposed　model　has　obvious　advantage　significantly　better　than　the　existing　models　in　the　cloudedge-end　cooperation　environments.　©　2022　IEEE.