A tiny logger to find which feature ate the OpenAI budget — and a $113k bill horror story
Pangji1003 · reddit · 2026-10-11
The author built a small logger to see which app feature was burning his OpenAI budget: after each call he logs token counts plus a feature name and prices them himself. Self-pricing turned out to be full of traps—cached reads bill differently (cache writes can cost more than normal input), reasoning tokens bill as output though you never see them, providers disagree on whether cached tokens sit inside the prompt count, and audio tokens can cost several times the text rate.
Digging deeper he found worse: a dev on Google's forum got €290 in search charges atop €10 of LLM usage with the spend cap never triggering; a timeout plus auto-retry on long-context calls quietly doubled someone's bill for weeks; one startup got a $113k AI bill in a single month.
His verdict: the code-gen feature ate 70% of his bill—yet by token count the summarizer was far bigger (129M vs 30M tokens) at only $14 total. The logger grew into a product, LLMtrack.
Related event: Developer builds logging tool to trace OpenAI bill burning budget(2 posts)→
More from Venture
- Google Could 'Eat OpenAI's Lunch' by Bundling AI Into Search and Workspace — HossCo · 2026-10-11
- Manus raises over $500M led by Boyu and IDG after Meta's $2B acquisition fell through — emmanuelvivier · 2026-10-11
- W1 real-time video model costs $0.003/sec, about 77x cheaper than Seedance — LearnWithBishal · 2026-10-11
- Lagos Life hits 10.15M players in 11 days, with 2.3M active today — Fowe · 2026-10-11
- PlayCanvas snake game shipped as a Snapchat Lens hits 1.6M plays in weeks — willeastcott · 2026-10-11
- Rumor: Anthropic's Fable 5.5 to land weeks ahead of its IPO, claims big leap — kimmonismus · 2026-10-11