Unexpected Codex Credit Usage – 17.8M Tokens in 12 Turns

Hi,

I’m trying to understand whether my Codex usage is normal or if there may be an issue with the usage reporting.

What happened:

  • Purchased 500 Codex credits (about $20).
  • Used the VS Code Codex extension for roughly 30 minutes.
  • Made edits to only around 7–8 files.
  • One file was about 1,000 lines, the others were much smaller.
  • I was mainly asking Codex to modify existing files, not generate large amounts of new code.
  • Auto-refill was disabled but credit become minus 442 .
  • In past when using plus subscription i could edit more than this .

My usage dashboard shows:

  • 12 turns
  • 17.8 million tokens
  • Lines of code: 0
  • Credit balance eventually went negative - 17$

What confuses me is that 17.8M tokens across only 12 turns works out to roughly 1.5M tokens per turn. That seems extremely high for the amount of work performed.

I also experienced GPT-5.5 errors during the session, and after that I only made a very small number of additional edits.

Questions:

  1. Is this level of token usage expected for a small coding session in VS Code?
  2. Does Codex repeatedly re-read the entire workspace on every turn?
  3. Can failed requests or retries consume large amounts of tokens?
  4. Is there a way to view token usage per turn or per request?
  5. Has anyone else seen similar usage numbers with relatively small edits?

Any help would be appreciated. I have screenshots of the usage dashboard if needed.

Thanks.

I see no reply from anyone and even support is not replying about usage .

I need help .

Hi @janak what is the support case id

Hello ,

My support case id is 10445189 .

Hi @janak,

Sorry for the delayed reply, and thanks for sharing the details and case number.

For context, a separate Codex usage-limit incident was investigated on June 26 and resolved on June 29. It involved incorrect rate limiting for some accounts; the cause of the token total and negative balance in your report remains unconfirmed.

Since there have been no further updates here, I’ll close this thread. If you’re still seeing unexpected usage now, you’re welcome to start a new thread with recent timestamps and timezone, your model and reasoning setting, and a screenshot with account and billing details hidden.

~ AD