My side project's API bill came out to 30,000 won: is that normal?
I'm building a side project on my own, and this month I got charged about 30,000 won for the OpenAI API. I only have a few friends using it—does that seem a bit off?
I checked the logs and found a part where the same request is being sent repeatedly, but not adding caching seems like the biggest issue. How much do you all spend on API costs for personal projects?
10 answers
30,000 won for a few friends is a bit weird lol. Start by checking the logs.
Do you have a source? I don’t get how 30,000 won can be normal when there are only a few users.
Agreed, caching is 80% of it.
You already knew caching wasn't set up... That's the answer. If you throw it into Redis using the same prompt hash, costs drop dramatically. On my side project, it comes out to about 3,000 KRW a month, and most of that is from testing during development—real usage is basically 0 KRW.
But isn't that outdated? These days, with gpt-4o-mini, that volume wouldn't even cost 30,000 won. I'm more curious about which model you're using.
I had the same issue. I was firing off full prompts every time in my dev environment, but when I checked the logs, I saw retries running 200 times a day. The retry logic had an infinite loop with no backoff... Just fixing that cut it by 90%. If you break it down by model in the usage tab of the OpenAI dashboard, the culprit shows up right away.
What worries me more than the cost is that if it’s 30,000 won for just a handful of users, the per-request cost doesn’t add up... Could your account actually have been compromised? Please check whether your API key has been exposed on the client side.
I just set a $5 monthly cap and use it that way. If I go over, it dies on its own, so I don’t have to worry.
At 30,000 won, it’s a bit much for a side project, but as users grow, it’s a cost you’ll have to handle anyway, so it’s better to set up the structure in advance. Just put these three in place: caching + rate limiting + routing to smaller models.
Well, I think the first step is to look at why the same requests are being repeated before adding caching. If you fix that, it might be enough even without caching.