I’ve been using Cursor regularly, but lately the costs feel much harder to understand.
Some days I see sudden usage spikes, even though my workflow hasn’t changed much. I know Agent mode and large context can use more tokens, but the jumps I’m seeing don’t always make sense to me.
I’m not trying to complain for the sake of it. I just want to understand what is happening. It would be really helpful if the dashboard showed a clearer breakdown of what caused each spike, like the model used, files read, context size, and token usage per request.
Right now, it feels hard to connect what I did in the editor with the usage shown in the dashboard, and that makes the pricing feel a bit unpredictable.
This is fair feedback. Agent mode can consume significantly more tokens than you’d expect because each step may re-read files, expand context, and issue tool calls that all count toward usage. Long sessions compound this since conversation history grows with every turn.
Right now, the usage page at cursor.com/settings shows a daily breakdown by model and category (included vs. on-demand), but it doesn’t provide per-request granularity — that’s the gap you’re describing.
A few practical things that may help in the meantime:
-
Start fresh chats for new tasks instead of continuing long sessions. Each turn carries forward the full conversation history, which inflates subsequent request costs.
-
Check your model selection. Max mode and premium models consume tokens faster. Auto mode picks the most appropriate model for each task, which often reduces token spend.
For the dashboard improvements you’re describing (model used, context size, token cost per request), this has been discussed in detail here: Context Window Inspector & Agent Usage Profiler. Voting on that thread helps our product team prioritize it.