Codex Pro 5x quota drops 2-3% for a single normal prompt, equivalent to 10-15% of Plus 5-hour quota

Open 💬 1 comment Opened Jul 2, 2026 by waitingformywind

Summary

I am on Codex Pro 5x. Recently, a single normal prompt can consume about 2-3% of my Pro 5x quota.

Since Pro 5x is 5x a Plus quota, this is equivalent to roughly 10-15% of a Plus 5-hour quota for one prompt. That implies a Plus user could only send about 6-10 normal prompts per 5-hour window, which seems clearly abnormal.

This does not look like normal high usage. It looks like a quota accounting / token burn issue.

Observed behavior

  • Plan: Codex Pro 5x
  • Date observed: 2026-07-02
  • A single normal prompt consumed approximately 2-3% of Pro 5x quota
  • Equivalent Plus 5-hour usage: approximately 10-15%
  • At that rate, one Plus 5-hour quota would only support about 6-10 normal prompts

Expected behavior

A single normal prompt should not consume 2-3% of a Pro 5x quota.

Even if the prompt has some context, the effective cost should not make a Plus-equivalent 5-hour quota disappear after only a few ordinary prompts.

Possible related areas to investigate

Please check:

  • server-side quota accounting / rate_limits.used_percent
  • repeated context billing
  • background agents or unfinished worker sessions
  • ambient suggestions / background summaries
  • subagent or forked thread sessions
  • usage meter display vs actual server-side metering
  • long-thread compaction or replay causing excessive billing

Related public reports

This appears similar to recent reports in this repository:

  • #30034 - Codex Pro 5h quota drops 2% after trivial prompts and capacity errors
  • #30785 - Usage draining much faster than yesterday
  • #30784 - Credit consumes too fast
  • #29963 - Codex quota consumption has a serious bug
  • #29955 - Quota drained instantly: 100 credits gone after 1 message

Impact

This makes Codex very difficult to use reliably, because even short conversations can consume a large fraction of the quota.

View original on GitHub ↗

This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗