The charge is by the cluster and does not depend on what is running on it. You may be running a python notebook on 1 driver + 5 nodes, but you will still be charged for the 6 nodes that are running. If you want to reduce the code you will put such tasks to use a separate cluster but on the flip side, it will take time for each individual cluster to spin up. So the best way would probably be to have all your tasks run on the same job cluster config with autoscale on.