I think an important point is getting lost in parts of this discussion: this is not about expecting unlimited Codex usage from a Pro subscription.
I fully understand that Codex is computationally expensive, that different models and reasoning levels consume different amounts of resources, and that a subscription needs to have limits.
The problem is transparency and predictability.
It starts with the way the plans themselves are described. Pro is advertised as providing 5x or 20x the Codex usage of Plus. But 5x or 20x of what, exactly?
If the underlying Plus allowance is not expressed in a stable, understandable unit, then a relative multiplier does not tell customers how much usable capacity they are actually buying. And even the published model-dependent usage ranges are so broad that they are of limited help when planning real work.
More importantly, once we start using Codex, we still cannot meaningfully see why our allowance is being consumed at the rate it is.
Many people in this thread are describing essentially the same experience: workflows that we were able to run for weeks without coming close to our weekly limit are now consuming a very substantial percentage of the allowance, sometimes even for relatively simple tasks. As a Pro user, I can now literally watch percentage points disappear during tasks that previously would barely have been noticeable.
That does not necessarily mean OpenAI intentionally reduced the limits. There may be changes in model behavior, context handling, reasoning, tool calls, compaction, metering, or something else entirely.
But from the users perspective, without transparent accounting, all of those possibilities look exactly the same: the effective amount of work we can do with the subscription has suddenly decreased.
That is why I think what several people here have already requested would make such a big difference. At minimum, users should be able to see:
- the actual consumption of each completed task
- how much was deducted from the 5-hour and weekly allowances
- which model/reasoning configuration caused that consumption
- a usable history of consumption;
It does not have to be perfectly predictable down to the last token. Even approximate accounting would be a huge improvement. What matters is giving customers enough information to understand their usage and make informed decisions about how to spend their allowance.
The new option to purchase weekly resets or additional usage is useful as an emergency valve, but I don’t think it solves the underlying problem. If I don’t know what consumed my existing allowance, buying another reset simply gives me another opaque bucket that may disappear just as unpredictably. For a professional workflow, that is very difficult to budget around.
There is another reason why this currently feels uncomfortable. Tibo wrote publicly on August 21:
“We’ve investigated a few messages about codex usage limits being different. That’s not something we change without engaging the community and being transparent.”
I appreciated that statement, and I still take it at face value. But the current experience described by many users in this thread feels indistinguishable from a substantial reduction in effective usage.
That is exactly why more visibility would help both users and OpenAI. If the limits themselves have not changed, transparent per-task accounting would make that much easier to demonstrate. It would also allow us to identify whether a specific model, setting, tool, repository, or workflow is responsible for unexpectedly high consumption instead of everyone having to speculate.
I like Codex a lot. That is precisely why this matters. People are trying to integrate it into serious development workflows and are willing to pay for substantial usage. We are not asking for unlimited compute.
We are asking to know what we are buying, what we are consuming, and how to plan around it.
.
