Three support cases and a month later, OpenAI has not returned to me with further findings. I downgraded from Pro to Plus on September 16, lost the better part of a week, and received neither a ruling nor a remedy. Their official language seemed to imply they had no documentation or standard to act on. Except that wasn’t even remotely the end of the problem.

The public reports go back much further. A GitHub issue opened last December describes the weekly allowance immediately disappearing on a Pro-to-Plus downgrade. It remains open. Another still-open report, filed in August, received an October 17 complaint about paying and having six days to wait. This isn’t a one-off freak accident. There is a recurring issue with Codex usage after a plan change, and OpenAI has neither dismissed these reports nor published conclusive findings.

To their credit, OpenAI has acknowledged an allowance error before. On October 20, Tibo said approximately 40,000 accounts had limits inconsistent with their selected plans, following reports of unexpectedly fast usage drain. He said the issue was corrected, affected users were being contacted individually, and they would receive credits. The tweet was quoted in the GitHub discussion, but whether these downgrade-related issues were included in the 40,000 remains unclear. I found no public documentation explaining how the error occurred or how remedies were determined for each subscription tier. This could have been a support document for customers to look up.

In one GitHub account of an October correction, the author reproduced an email explaining that a higher Pro allowance had mistakenly survived a downgrade. When OpenAI corrected the allowance, usage already accumulated that week counted against the lower limit. OpenAI said it had added 12,500 credits, worth $500. The recipient had not independently verified the balance, but the reported response at least supplied an explanation and a remedy. My support cases produced neither.

Elsewhere, a subscriber reported a partial refund after a premature downgrade, while others reported corrective usage credits. Different faults can warrant different remedies. What I cannot find is an intelligible standard connecting the fault, the investigation, and the outcome. OpenAI has tools to help organize reports and communicate decisions — they make and sell them. My three cases give me no confidence that a coherent system, or a consistent policy for correcting it, governs the process.

What remains of a reset when the company cannot explain the allowance around it? Mine began as less than $5 lost from a $20 plan. OpenAI could have settled it quickly or completed the review. Instead, the customer is left to start the process again. A fourth case should start with OpenAI’s own internal audit, not another request from the aggrieved customer.