I am on ultra and I feel like token usage is INSANE. I am going to run out halfway through the month coding for a few hours a day. Anyone else have this? For this amount of money and the crazy use rates, combined with how bad Auto is this is likely my last month on cursor.
7 Likes
SAME! Like holy â– â– â– â– its â– â– â– â– â– â– â– terrible.
i personally noticed, even if i just say “hi” to the agent, having about 3 sentences in my rules, the token usage is around 10 thousand
2 Likes
If auto wasn’t a completely worthless it would be a little easier to stomach but this is next level. I did a few minor bug fixes today and used up ridiculous tokens, even refining the hell out of my prompts.
1 Like
I reported a glitch the other day (when the ClaudeSonnet4 agent outputted its ENTIRE input + system prompt) and it literally had EVERY MEMORY EVER MADE in the prompt between
Do you have memories enabled as well?
1 Like
Yeah I have memories enabled - does that contribute to this extreme usage?
Judging by the leaked output of the agent’s inputs the other day - it seems like this ABSOLUTELY contributes to it. My memories along were around 4k tokens!
1 Like
That’s an absolute joke. I just turned off all memories. Let’s hope that tones down this token usage a bit. I feel completely ripped off - does anyone from cursor ever reply on these forums?
2 Likes
@deanrie does but its difficult when the people responsible for fixing the problems and answering support emails completely ghost us. I’m considering charging back on my credit card atp after being completely ghosted by support after being promised a refund. This is kind of unreal.
1 Like
Some dude named Michael always reaches out then never responds. The token usage is still off the charts for simple fixes… We can’t be the only people seeing this. Are there any good alternatives to cursor?
2 Likes
No good alternatives. Cursor is best in class at the moment. Its been terrible for me since July 4-5th honestly. I’ll see how it plays out!
1 Like
I am Pro+. Cursor tell me, that my quota will ends nearly in the end of my third week of the billing period.
1 Like
It’s insane. I am desperately searching for an alternative as I am in the middle of a project and this pricing/token usage is flat out the worst i’ve ever seen. Going to demo augment.
1 Like
Premium (Ultra users)
- Claude 4 or Gemini 2.5 Pro for quick edits and analysis.
- Grok 4 MAX for everything else.
I am on ultra and simply using Claude 4 EATS through tokens like no tomorrow. I don’t see how your guide makes any sense.
Gemini is cheaper and smarter. But he has frequent problems with edit_tool.
Maybe I shouldn’t have combined them into one bullet… 
| Kind | Model | Input | Output | Cache Write | Cache Read | Cost |
|---|---|---|---|---|---|---|
| Unknown | gemini-2.5-pro-preview-05-06 | 22,736 | 806 | 0 | 169,579 | Included |
| Unknown | gemini-2.5-pro-preview-05-06 | 30,288 | 792 | 0 | 159,654 | Included |
| Unknown | o3 | 17,084 | 1,190 | 0 | 21,551 | Included |
| Unknown | o3 | 5,939 | 927 | 0 | 96,909 | Included |
it doesnt seem cheaper than o3?
o3 is 30% more expensive if I remember correctly. And she’s dumber.
Compare the number of tokens and the cost. And I don’t use Claude-4-Thinking.
1 Like
I see. You’re saying actual cost compared to token usage. I’ll give Gemini a go.
I still can’t get on board with using Auto. It’s really bad. Like gpt 1.0
1 Like
I’m currently using it to find out how something works in the repository. It is good because of the semantic search tool.