Anyone else watching their LLM API bill climb for no good reason?
If you’re shipping anything on GPT / Claude / Gemini, you’ve probably felt this:
Same product. Same quality. Bill goes up because prompts got longer — more instructions, more context, more “please carefully…”.
A big chunk of that spend is often pure filler tokens. Not smarter prompts. Just more words.
Curious for people actually feeling this in production:
Roughly what % of your monthly AI budget is input tokens vs output?
Have you tried any prompt compression / cleanup — or is it still “just write shorter”?
What’s the highest monthly LLM bill you’ve hit so far?
Building CuToken (cutoken.in) around this exact pain: compress the prompt, keep the intent, cut the waste.
Drop your numbers / stack in the comments — useful to see how common this is beyond hobby usage.
The cat-and-room concept gives the habit tracking a very different feel from the usual productivity apps. The 10-day challenge origin story is interesting too.