Hi,
I think it would be great if simple questions like CMD+K or Ask take less credits than the Agent.
Maybe I am missing something, but it looks like the Agent is taking a lot more tokens than CMD+K or Ask but they are using the same amount of credits.
I can understand that it’s a bit tricky to evaluate, but a sort of rule of thumb like Agent takes 2-3x more tokens could be helpful.
Thanks!
You can use in CMD K and Ask a less capable model (non-premium, so it would cost less right?)
But overall if you select the same model in all cases it would cost the same e.g. 1 request or depending on model
You can see here how much each model costs
Some extra capable models like MAX cost per request and per tool call as described there. Regular models cost 1 request. some smaller 1/3 request and so on.
Cursor team has listed details on that page.
Some models are also free so you can use those to ask quick questions in Ask
For sure! I’ve been using Deepseek V3 a lot since it’s the best out of the non-premium models in my opinion. I’ve used a lot gpt-4o-mini as well in the past.
My point was more related to CMD-K and Ask requests are usually smaller than in Agent or Edit, so the number of tokens used would be less (since they are usually smaller requests), that’s why I was suggesting to maybe use less than 1 credit for these requests. Because if we compare it to Agent, I would assume that agent takes 2-3x more tokens.
That thought came from the fact that if I ask a simple question to Sonnet like “Build an HTML page” using CMD+K or Ask versus Agent, usually the Agent will go beyond and might take more tokens, that’s my assumption.
Does that make sense?
Cursor doesnt charge by tokens. they list requests or USD
perhaps to explain why they likely do that:
- Agent needs not just a prompt but there are several AI requests in between that evaluate what the model must do and what to add to context, which tools to call etc. which means there are several requests to AI
I know, but whether we want it or not, if you ask a simple request (CMD+K or Ask) to Claude vs a complex one (Agent), it could be nice to use less token, but it was just a suggestion, maybe it makes no sense.
yeah, i updated my answer above in the meantime why its not by token, that means you cant compare with other AIs just so simply
1 Like
True, I much prefer the credit usage than token usage, it’s indeed much simplier. I just thought it would be great if the credit is a bit more dynamic than 1 for 1.
1 Like