The first reaction to opening the site is almost unavoidable:
This is a fan site.
Usage-limit resets are normally the least glamorous part of a SaaS product. Yet Codex and ChatGPT Work have produced something stranger: OpenAI’s Tibo Sottiaux posts public reset announcements, third-party sites monitor them, and the community has built histories, interval statistics, alerts, an API, and even an MCP endpoint around the ritual.
There is even a 🙏 beg control on the page.
This is no longer merely a quota dashboard. It is an observatory pointed at Tibo’s RESET button.
The headline snapshot on September 10, 2026 was striking: a public-feed mirror showed 53 recorded reset-related events, a 6.9-day average interval, and a 67.7-day longest wait. But those 53 events do not mean that everybody received 53 immediate full resets. The history mixes banked resets, direct automatic/global resets, and compensation targeted at affected users.[1][2][3][4]
That distinction is exactly why a heavy user waiting to move to the 20x tier can end up thinking:
A reset might land. Please let me upgrade first. I am literally trying to pay you.
Odd sentence. Surprisingly coherent economics.
1. A fan site for a quota button somehow became infrastructure
Codex Resets states its purpose plainly: it watches Tibo’s public Codex reset announcements so users do not have to.[1]
The site offers history, a reset-probability estimate, Telegram and email alerts, and public API/MCP access. MCP is a common way for AI agents to call external tools and services. So we have reached the point where an AI can ask another service whether the human responsible for giving AI users more AI quota has pressed the button again.
The recursion is excellent.
The site clearly says it is not affiliated with OpenAI, explains that a robot classifies the public posts, and credits wong2 as the builder.[1]
In one crawl snapshot, its sponsorship panel also advertised roughly 150,000 monthly visitors and a $699-per-month ad slot. That visitor figure is the site’s own changing snapshot, not an independently audited metric, but the business model is still comedy gold:
A site that tracks free quota resets acquired a paid advertising product.
Someone built a Bloomberg terminal next to the Tibo central bank.
2. 53 events, 6.9 days average, 67.7 days longest: impressive numbers, not a timetable
A feed-synchronized view checked on September 10 showed 53 events, an average interval of 6.9 days, and a longest wait of 67.7 days.[2]
A search snapshot from September 8 still showed 52 events and a 7.0-day average, with the same 67.7-day maximum.[1]
That is not necessarily a contradiction. It is what a live aggregate looks like when new events are added.
The dangerous mistake is reading 6.9 days as a schedule.
It does not mean Tibo reliably appears every 6.9 days. Resets have been associated with outages, product launches, user milestones, promotional events, and other ad hoc reasons. The 67.7-day longest gap is the giant warning label on the average.[1][2]
Putting “next reset in about a week” into a calendar because the mean says 6.9 days is like using annual rainfall to predict the exact hour of next Tuesday’s shower.
Useful context. Terrible appointment system.
3. The big trap: 53 events are not 53 universal instant resets
The word “reset” covers several mechanically different things.
| Type | What happens | User action | How to read it in the total |
|---|---|---|---|
| automatic / global / direct reset | Eligible usage limits named in the announcement are refreshed directly | Usually none | Closest to “the tank refilled by itself” |
| banked reset | A one-time reset is saved to the account | User chooses when to apply it | Does not necessarily refill usage immediately |
| targeted compensation | A reset is granted to a specific affected cohort | Depends on the offer | Must not be treated as universal |
OpenAI’s own documentation explicitly separates banked resets from automatic/global resets.[3][4]
A full banked reset can refresh the eligible five-hour and weekly windows and change the weekly reset date. If there is nothing eligible to refresh, it is not consumed. An automatic/global reset, by contrast, is applied directly and does not create a saved reset.[3][4]
So the clean mental model is:
Banked = ticket. Global = refuel now.
OpenAI’s September 2026 guidance lists banked resets for eligible users on September 3 and 4 and an automatic reset on September 7.[4] Public trackers also recorded a September 5 banked-reset announcement and a direct reset timestamped September 8 UTC.[1]
On September 9, OpenAI Status also reported an incident involving unexpected Codex usage-limit resets. A subsequent public-feed item described another reset for users whose banked reset had not fully applied.[2][5]
That is precisely why “53” is a great measure of reset culture, but not evidence of 53 universal free full tanks.
4. A banked reset is a saved reset ticket, not credit and not an automatic reset
The simplest description of a banked reset is: one saved right to refresh your eligible quota later.[3]
OpenAI says it is not purchased credit and does not permanently increase the plan’s allowance. The user applies it from the usage interface, and it is consumed only when it successfully refreshes at least one eligible window.[3]
This is operationally different from Tibo announcing a global reset.
- Everyone in the named eligible scope gets refreshed directly: automatic/global.
- You hold a one-use reset for later: banked.
- You buy extra usage or an instant reset: yet another mechanism.
Putting all three under the casual word “reset” is how a fun statistic becomes a wrong financial assumption.
The bank does not contain cash. It contains one coupon that says refill the tank.
5. Why a 20x user waiting for an upgrade starts caring about reset timing
OpenAI currently describes Pro $100 as offering 5x Plus usage and Pro $200 as offering 20x Plus usage.[6]
That means the step from 5x to 20x is a 4x increase in allowance, not a twentyfold increase over 5x.
OpenAI also says that when an upgrade succeeds, the new limits apply immediately.[6]
Now consider somebody already exhausting the 5x tier with Work or Codex. For that person, 20x is not decorative headroom. It is capacity for work that already exists.
If the upgrade is stuck while another global reset appears likely, the user naturally thinks:
If I were already on 20x, I could potentially resume with the larger allowance.
That does not make future resets guaranteed. Eligibility, plan scope, region, account status, and the exact limits affected can vary by announcement.[3][4]
So the defensible purchasing logic is not:
Tibo averages 6.9 days, therefore 20x pays for itself.
It is:
I would use 20x even if Tibo never resets again. Any reset is upside.
Do not put “Tibo expected value” in the fixed-income column of your budget.
6. After almost a day of waiting, today’s free time can matter more than the monthly price
The conversation behind this article included an anonymized example of someone who wanted to upgrade but had an individual plan-change issue stuck for nearly a day.
For a heavy AI user, the obvious cost is not necessarily the subscription price. It is perishable working time.
Astra and Codex are most valuable when you can run a long exploration, inspect the result, redirect it, fix the environment, and convert the good parts into automation. Six free hours today are not interchangeable with six hypothetical free hours next week.
So “just wait” can still have an opportunity cost even when no money has been lost.
And the situation is inherently funny because it is not a reluctant buyer asking for a discount. It is:
I will pay. I want the larger plan. Please sell it to me.
At that point the desired button is not BANKED RESET.
It is UNBLOCK.
Ideally somebody presses that before Tibo presses RESET again.
7. No public-wide billing incident does not mean no individual account issue
OpenAI’s status history for September 9 showed an incident involving elevated errors for Pro/Plus conversations and another involving unexpected Codex usage-limit resets; both were marked resolved when checked on September 10.[5][7]
The public history checked at that time did not show a broadly labeled incident saying that Pro plan changes or payments were failing platform-wide.[7]
But that only supports a narrow statement: no such widespread incident was displayed on the public status page.
It does not prove that an individual billing, review, or plan-change case cannot exist. Status pages summarize service health; they are not public ledgers for every account-level support case.
A sunny weather map does not prove your doorbell works.
8. Why the Tibo shrine grew an API, MCP, alerts, and ads
The tracker ecosystem looks absurd until the incentives are listed.
- Work/Codex quota has real productive value.
- Bonus resets are discretionary rather than a fixed calendar event.
- Important signals often appear in Tibo’s public posts.
- “Reset” can mean global, banked, or targeted compensation.
- Time zones make “today” annoyingly ambiguous.
- Heavy users gain the most from learning about a refill quickly.
Given those conditions, someone builds a dashboard.
Then alerts appear.
Then an API appears.
Then MCP appears.
Then sponsorship appears.
Humanity built an observability stack around one person’s RESET button.
So yes, “fan site” is accurate. It is simply not the kind that collects photos of the idol.
It collects the idol’s quota-replenishment logs.
9. Conclusion: read 53 as a cultural statistic, not a promise
Five checks keep the tracker useful without turning it into superstition:
- Count: how many public reset-related events were captured?
- Type: direct/global, banked, or targeted compensation?
- Scope: which plans, regions, or users qualify?
- Time: UTC, Pacific Time, Japan time, or something else?
- Decision rule: would the plan still make sense if no free reset ever arrived again?
Fifty-three events, a 6.9-day average, and a 67.7-day maximum are objectively hilarious numbers for a quota-reset tracker.
But the more interesting story is the infrastructure that emerged around them: one public feed became a history, alerts, an API, MCP, and even an ad product.
It really is a fan site.
For someone waiting on 20x, the final rule is equally simple:
Buy 20x because you have enough real work to use it without assuming another reset. If a reset arrives, celebrate.
If you already decided to buy and the transaction itself is blocked, however, the prayer changes:
Tibo, thanks for the RESET. Now could somebody please press UNBLOCK? There is work to do.
References (7)
- Codex Resets, ‘Codex Limit Reset Tracker & History’ codex-resets.com
- AI Resets, Codex public-feed mirror / current aggregate snapshot ai-resets.com
- OpenAI Help Center, ‘How banked Codex resets work’ help.openai.com
- OpenAI Help Center, ‘Managing usage with GPT-6 Astra in Work and Codex’ help.openai.com
- OpenAI Status, ‘Investigating unexpected usage limit resets’ status.openai.com
- OpenAI Help Center, ‘About ChatGPT Pro tiers’ help.openai.com
- OpenAI Status, History status.openai.com
