OpenAI's head of Codex, Thibault Sottiaux — known in the community as Tibo — set a public deadline for his team on social media on October 4. Starting that day, for the next 28 days, the Codex team will either ship an update each day that clearly improves things for most users, or issue a full usage reset for everyone.
"Over the next 28 days, each day we'll either ship one thing that is a clear improvement and relevant for most codex/work users or ship a full reset."
Counting from October 4, the window runs through roughly November 1. The commitment covers both Codex and ChatGPT Work, which share a single usage pool.
Why "reset" is the bargaining chip
Among Codex users, "reset" carries real weight. Codex's paid plans are metered in five-hour windows plus a weekly cap, and heavy users often burn through their allowance in half a day — complaints about limits draining too fast have never really stopped on the community forums. Over the past few months, Sottiaux has repeatedly announced resets for all paying users on social media, and third-party developers have even built dedicated web pages to track whether he's about to announce another one.
In late August, an OpenAI staffer posted on the official developer community that Codex's weekly active users had just crossed 20 million. To mark it, every Codex and ChatGPT Work user got a "banked reset": instead of refreshing immediately, the allowance was held in reserve for the user to trigger whenever they chose. In the replies to that same post, one user reported burning through their allowance from 100% to 0% in six hours.
So this commitment gives users a safety net: even on a day without a real update, they still get a reset. For the team, it's a way of forcing a rhythm on themselves — if they can't ship an improvement, they have to make up for it with compute.
Doing the math on the 28 days
Taken literally, the promise produces 28 "outcomes" over 28 days, with improvements and resets adding up to that total. Here's a rough estimate: if the team ships an improvement on half the days and a reset on the other half, users would get 14 full resets over the four weeks — roughly one every two days. Given Codex's current five-hour-window-plus-weekly-cap structure, that would push heavy users' total available usage well above normal, though by how much depends on which tier of allowance each reset refreshes, something OpenAI hasn't specified.
On the flip side, if the team manages to ship an improvement every single day, users won't get a single reset — but they'll see 28 updates over four weeks instead. Which outcome is the better deal depends on whether you're someone who's limited by usage caps or by missing features.
Sottiaux hasn't specified who decides what counts as a "clear improvement" or "relevant to most users." Nor has he clarified whether a full reset includes the weekly cap. Those details will determine how much this 28-day promise is actually worth — for now, that can only be judged as OpenAI delivers on it day by day.
Sources: Thibault Sottiaux's public social media statements, OpenAI developer community announcement, CocoLoop, and OpenAI's community post for the 20-million active-user figure and the banked-reset mechanism; the 28-day window's start and end dates are estimated from the date of the statement.