openai.com
OpenAI Quota Refresh Signal Monitor
05/21/2026, 04:47:00 PM@Wei Wang
Alert: Codex rate limits reset — Sam Altman's 1-like dare played out in real time
On May 19, 2026, Sam Altman joked that one like on his tweet would make Tibo reset Codex rate limits — and it happened within three hours. Here's the full chain: the tweet, the confirmed reset from @thsottiaux, and the adjacent OpenAI Guaranteed Capacity announcement signaling broader compute scarcity.
⚡ Alert: Codex rate limits reset — confirmed by @thsottiaux, triggered by Sam Altman's public dare
On May 19, 2026, OpenAI Codex users got an unscheduled usage reset. The trigger: Sam Altman posted a joke on X, OpenAI's account retweeted it, and Tibo Sottiaux — the Codex product lead — followed through within hours.
This alert covers the full chain: the joke, the reset, and the adjacent Guaranteed Capacity announcement that adds longer-term context to why OpenAI is talking about quotas and compute access at all.
What happened
~18:33 UTC, May 19 — Sam Altman posted (and
@OpenAI retweeted):Loading content card…
The tweet is a classic Altman deadpan. "Tibo" refers to Tibo Sottiaux (
@thsottiaux), a Codex product manager at OpenAI who has become something of a community folk hero for manually resetting Codex usage limits during milestone weeks.~21:34 UTC, May 19 — Sottiaux confirmed the reset on his own account:
Loading content card…
"At some point I counted one like, so I went and did the thing. Enjoy your Codex usage. We had a few things to celebrate this week 👀"
If you're on a Codex paid plan, your weekly usage limits were reset at that point. The "a few things to celebrate" hint likely refers to the Guaranteed Capacity launch (see below) and possibly an undisclosed product update.
The pattern behind the joke
This is not the first Tibo reset. A community thread from April 28 shows Sottiaux explicitly resetting Codex rate limits for all paid plans to "celebrate" GPT-5.5's strong week 1, and the behavior dates back further. One user offered a useful clarification 2:
"They're not giving you more tokens per se — they're just saving you time. If your limits reset on the 28th, and your official limit reset would happen on the 5th of next month, then a Tibo reset moves that official reset earlier."
In other words: early resets compress the wait period, not the total allocation. Worth keeping in mind if you're planning heavy usage after a reset.
Also May 19: a separate (and more permanent) capacity signal from @sama
Within the same hour as Altman's joke tweet, he posted a substantive note about supply:
Loading content card…
"customers are increasingly asking us for certainty on capacity. as models get better, we expect that the world will be capacity-constrained for some time. we are offering discounted tokens for 1-3 year commits."
This directly accompanied the official OpenAI Guaranteed Capacity announcement 3:
Loading link preview…
Key terms: 1-3 year compute commitments, discount increases with commitment length, applicable across OpenAI's cloud provider portfolio and model families. This is for enterprise-tier customers planning critical AI workloads — not the same as the ad-hoc Codex developer resets above, but the two signals are connected: OpenAI is explicitly telling the market it expects sustained compute scarcity, which is why both the enterprise capacity product and the developer-goodwill resets matter as signals.
May 17 reset (the one before this one)
Worth logging: two days earlier, on May 17 at ~22:47 UTC, Sottiaux issued a different kind of reset after a bug 4:
"Codex usage limits are back to normal. For ~2 hours we had some issues where the enforced limits were wrong for the majority of users on a subscription. We waived usage consumption for that time and things should be set again to exactly to where you left it."
That one was remedial (a bug correction, not a bonus reset). The May 19 reset was additive.
What to watch next
Sottiaux's "👀" suggests something else is coming. His current engagement pattern — posting about Codex on weekends, hinting at upcoming features — has preceded product launches before. Keep watching
@thsottiaux and @OpenAIDevs for specifics.For enterprise users, the Guaranteed Capacity terms require contacting OpenAI sales directly; no self-serve option is currently published.
Sources: public X/Twitter timelines of @OpenAI, @sama, and @thsottiaux. All timestamps in UTC.
Add more perspectives or context around this Drop.