It was a Tuesday morning. I opened my Anthropic dashboard to check usage like I do every few days, and there it was: $80.17.
I stared at it for a solid ten seconds.
I'm a solo dev. I build small Mac apps. I do not have $80 floating around for a single month of API calls that I barely remember making.
How it happened
Here's the thing — I wasn't even building anything big. I had been iterating on a feature for one of my apps, running Claude back and forth to refine some logic, and testing a few prompts. Normal stuff. The kind of session where you think "this'll be like $3."
But I had left a loop running longer than I realized. A script that was calling Claude repeatedly for batch processing some test data. I forgot about it, went to bed, and woke up $80 lighter.
No alerts. No cap. No warning. Just a bill.
The problem nobody talks about
LLM billing is genuinely different from anything else in software.
With AWS, you at least get billing alarms. With Vercel, there are spending limits. But with most AI APIs, the default is "we'll let you spend whatever, good luck."
And the usage is invisible. You can't feel it. Every little test call costs something. Every loop iteration. Every "let me just try this one more variation." It adds up in ways that RAM usage never did, because RAM is just... RAM. It doesn't cost you per thought.
Token usage is the new RAM usage — except RAM was free after you bought the machine.
What I built to fix it
After that bill, I wanted something that would show me token usage in real time, right there in my face while I was working.
I built TokenBar — a macOS menu bar app that sits in your toolbar and shows you live token counts and estimated cost as you work. It connects to your Anthropic API key, watches your usage, and gives you a running total so you're never surprised again.
It's not a dashboard you have to open. It's always there, next to your clock, like a fuel gauge.
$5 lifetime. I built it for myself, then figured other solo devs and indie hackers probably have the same problem.
The lesson
If you're using LLM APIs regularly, you need some form of visibility. The billing happens fast and silently. I've talked to other devs who've had similar moments — some much worse than $80.
Don't wait for a surprise. Keep an eye on your usage the same way you'd keep an eye on any resource that costs real money.
And if you're on macOS and want the quick fix: check out tokenbar.site. It's the thing I wish existed before I learned this lesson the expensive way.
Have you had an unexpected AI bill? Curious how others are handling LLM cost monitoring — drop it in the comments.
Top comments (1)
I have seen this before, now I have not been billed by AI but this is because of the API(pay-as-you-go) where the more total prompts you use the more you burn through these tokens and because it counts total this is why you got billed 80$ and if your claude model remembers your previous chats it will use up all the tokens from your previous chat if it retreives all the data