Consumption may vary depending on the type of activity. A voice session, a long correction, and generated audio do not have the same operational weight.