I am experiencing what appears to be the same issue, and the usage depletion has become significantly worse over the past few days.
My Codex weekly allowance is now being consumed roughly 10–20× faster than during my previous normal usage, despite working on similarly scoped tasks.
For one direct comparison, essentially the same implementation task consumed:
- About 25% of my Codex weekly limit using Sol at medium reasoning
- About 3% of my Claude Code weekly limit using Opus at medium reasoning, although it consumed around 65% of Claude’s separate five-hour limit
This is especially difficult to explain because, on paper, Sol is supposed to be cheaper to run than Opus. Yet for comparable work, Codex consumed roughly eight times as much of the available weekly allowance.
More recently, three small and tightly bounded code-review feedback tasks consumed approximately 43% of my Codex weekly allowance before the celebratory reset announced by Tibo for reaching 8 million active users across Codex and ChatGPT Work. After that reset, the remaining tasks consumed another 82% of the newly restored weekly allowance.
These were narrow review-feedback fixes, not repository-wide implementations or open-ended tasks. Consuming most of an entire weekly allowance for this amount of work seems completely unreasonable, even for a $20 subscription, and is far outside my previous Codex experience.
OpenAI has also been providing frequent usage resets as Codex reaches user milestones. While the resets are appreciated, they do not address the underlying problem and may actually hide how severely the normal weekly allowance is being depleted.
The removal of the five-hour limit may hide the issue further for many users. There is no longer an obvious short-window limit being exhausted, but at the current depletion rate, the weekly limit effectively feels like the old five-hour limit. Users may only notice the problem after most of their entire weekly allowance disappears within one working session.
My concern is that these resets are temporarily masking either:
- A serious metering or accounting problem; or
- A substantial intentional reduction in effective usage limits.
If the lower effective allowance is intentional, OpenAI should communicate this clearly. Otherwise, frequent resets risk gradually getting users accustomed to dramatically lower normal limits without explaining what changed.
I reported the issue to OpenAI support and eventually got the support AI assistant to escalate it. It has now been more than 48 hours, but I have received no human response, escalation reference number, or confirmation that the underlying usage logs are being reviewed.
Without per-session details for input tokens, cached input, output tokens, retries, tool calls, repository scans, and subagent activity, it is impossible to determine whether this is legitimate consumption or incorrect metering.
Given the sudden change, the scale of the depletion, the comparison with Claude Code Opus, and the fact that Sol should theoretically be cheaper than Opus, this does not appear to be normal model-cost variation.
I would also like to hear from other users about whether they are seeing the same pattern in either their weekly allowance or the previous five-hour allowance.
I noticed this change quickly because I carefully ration my usage, keep tasks tightly scoped, and try to get the most value possible from the $20 subscription. Under normal conditions, both the five-hour and weekly allowance decreases were reasonably predictable based on the size and complexity of the task.
That is why this recent change is so obvious to me. The consumption is not just slightly higher than usual; it is a sudden and very large increase that is completely out of line with my previous usage patterns. Since the five-hour limit was removed, the weekly allowance now appears to be depleting at a rate that feels comparable to the old five-hour limit.
Please share whether your weekly allowance is also dropping much faster than expected, whether you observed similar behavior in the former five-hour allowance, or whether your usage still appears normal.