Daily 'Improvement' or 'Reset for Everyone': OpenAI's Self-Imposed 28-Day Public Mandate for Codex
OpenAI's Codex lead pledged that for 28 days the team will either ship one improvement that helps most users every day or reset everyone's usage limits. Day one boosted generation speed by about 1.5x. We lay out the aims—and the criticism that it turns usage into a gamble.
The Codex Lead's Pledge: "We'll Ship One or the Other Every Day"
Thibault ("Tibo") Sottiaux, who leads the development of Codex and ChatGPT Work at OpenAI, made an unusual public commitment on X in early October. For the next 28 days, the team will either ship one thing that is "a clear improvement for most Codex/Work users" every single day, or else hand every subscriber a "full reset" of their usage limits.
The period runs for 28 days, from October 5 to November 1, 2026. There is no middle ground. There are no partial points for shipping something that doesn't help much—each day forces a binary choice between a valuable improvement and a reset.
The Daily Choice and Four Narrowed Focus Areas
The structure of the pledge is simple. Each day, the team picks one of the following:
- Ship one improvement: it must be a clear step forward that benefits many users.
- Hand out a full reset: on days when an improvement isn't ready, reset every subscriber's usage limits.
Tibo also said the team would narrow its work during this period to four areas.
- Product simplifications
- Efficiency gains that let you do more within the same allotment
- Breakthrough new features
- New models
Day One: Generation Speed Up ~1.5x, Reaching External Tools Within Hours
What shipped on day one (October 5) was a boost to the model's generation speed. The default GPT-6 Astra and GPT-6.1 Sol reportedly saw their output speed rise by roughly 50%, improving from around 30 tokens per second to roughly 50 via subscription access.
This is said to come from backend optimizations rather than a swap of the models themselves. The speed-up was reported to have propagated within hours to external tools that integrate via "Sign in with ChatGPT" (such as OpenCode, Pi, Amp, and Devin).
On subsequent days, multiple outlets reported, there were days when several small improvements shipped together, as well as days when a reset was actually handed out because the improvements fell short. The day-by-day specifics are recorded inconsistently across observers, so here we stay with the broad outline we could confirm.
Why Set a Quota in Public?
Behind this lies growing dissatisfaction over usage allotments. Criticism of reduced limits and the effective shrinking of higher-tier plans had continued, compounded by rivals such as Claude Opus 5.5, which showed an edge in benchmarks.
A development lead imposing a deadline on himself and publishing progress every day—this approach appears aimed at responding to frustration over stagnation in a "visible" way. It can also be read as a kind of performance of accountability, publicly committing development priorities to outsiders.
What It Means for Developers and Operations
For teams that use Codex daily, there are short-term practical gains. Faster generation shortens the wait time for agents and increases the number of iterations you can run in a day. Even on days without an improvement, a reset restores your allotment, so the design keeps immediate losses small whichever way things go.
On the other hand, daily updates create a cost of keeping up on the operations side. Because behavior and limits may change from day to day, it's worth noting that internal procedures and cost estimates become harder to pin down.
Concerns Behind the Convenience
This pledge has drawn criticism as well. Some point out that a design that hands out usage-limit resets as a daily "reward" treats the user experience like gambling. The view is that a reset is only a temporary relief and doesn't solve the structural problem of allotments being insufficient in the first place.
There is also concern that self-pressuring development—committing to ship something every single day—carries risks for quality and stability. The more you prioritize shipping on a short cycle, the more the time available for verification tends to be squeezed. Both of these remain at the "such concerns are also being raised" stage, and an assessment awaits the end of the 28 days.
References: 28 days of Shipping at OpenAI (OpenAI Developer Community), The New Stack, Nerdschalk, ExplainX
