Codex Resets
Codex Resets
A site now tracks every single time OpenAI's Codex boss resets usage limits on X. 35 resets so far — turns out 'generosity' is often just a bug report in disguise.
Based on reporting by Codex Resets — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Somebody built a tracker that does one very specific, very funny thing: it watches Thibault Sottiaux's X account and logs every time he announces a reset of Codex and ChatGPT Work usage limits. The count sits at 35 resets. The average gap between them is 8.9 days. The longest stretch without one — a drought, in the tracker's own gleeful phrasing — ran 67.7 days.
Read enough of these tweets back to back and a pattern emerges that Sottiaux never quite states outright. Some resets are genuine celebrations: Codex crossing 3 million weekly users, then 7 million, then 8 million, then 9 million active users across Codex and ChatGPT Work, each milestone marked with a reset like a company throwing open the bar. The $100 tier launch got one. The 1-year Codex anniversary got one, framed as Codex resetting its own limits, because apparently even the product has a sense of humor about this.
But plenty of the other resets are quieter admissions that something broke. A cyber-risk filter wrongly blocking requests. A caching optimization that drained limits faster than intended and had to be rolled back. Three separate reliability incidents in 24 hours. A rate-limit bug that skipped a promotional 2x boost for roughly 9% of Plus and Pro users. Each one gets the same treatment: an apology, a technical one-liner, and a reset, dressed in language that makes an outage sound like a gift.
There's also a running bit about banked resets — users stockpiling up to three unused resets, applying them whenever they like, sometimes getting a double reset stacked on top for good measure. During an internal event OpenAI apparently nicknamed 'RESET week,' Sottiaux even joked about the coincidence while mid-incident. The holidays got their own move too: a straight 2x usage multiplier through New Year's, sold as a gift rather than a limit change.
What the tracker really captures, without meaning to be that serious about it, is a live diary of how fast Codex has scaled and how often the infrastructure has had to catch up. Nine million active users in roughly a year is the kind of growth that breaks caching systems and billing pipelines, not just marketing decks.
My take — AI-written commentary, not fact-checked reporting
I like this tracker more than most product launches this month, because it accidentally does what OpenAI's actual status page never does: show you exactly how often 'we appreciate you' means 'we messed up.' Framing outages as gifts is a very Silicon Valley move, and I get why it works on users who just want their limits back — but somebody logging the pattern in public, unprompted, is a healthier form of transparency than anything in the official comms.
Read more about this at: Codex Resets