Image: bendrape / reddit.com · The maker’s posted cost dashboard showing US$104.53 in month-to-date spending
A dictation prompt test unexpectedly ran up a $104.53 bill
The maker reports that Claude Code expanded a dictation-cleanup test to Opus 5.5 at high effort without asking, costing US$104.53.
Credits
Earliest dated source: Details reviewed:
Set a hard API spending limit and require approval before a test changes model, effort or call volume.
The full story
The maker had built a dictation app that used Haiku to remove filler words and correct transcription mistakes. They asked Claude Code to try alternative system prompts against results that already existed.
According to the post, Claude decided to add a comparison against a stronger model and ran Opus 5.5 at high effort without asking first. The maker reports US$104.53 in spend from the expanded test. The post describes the unexpected model choice and cost; it does not share a billing audit or a public version of the dictation app.
What Claude did
- Expanded dictation-cleanup prompt evaluation to a stronger model without the maker requesting that comparisonClaude Code · Opus 5.5 (high effort) Source ↗
How it was done
Tools
Themes
Loading themes…
Sources
From a short post by the maker, so the details are as they tell them.
- Reddit: Opus 5.5 just spent $104.53 without me telling it to.
The maker's Reddit post describing the project and reported results.