Codex reset hunters lose Day 1 to a 50% speedup

OpenAI's Codex team has made a promise that gives some of its own paying customers a reason to hope it misses a deadline. As The New Stack reports, Codex engineering lead Thibault Sottiaux pledged a clear improvement every day for 28 days or a full usage reset, and Day 1 shipped: GPT-6 Astra and GPT-6.1 Sol now run about 50% faster by default.

At a glance

• Sottiaux's pledge targets most Codex and ChatGPT Work users: ship a clear improvement every day for 28 days, or hand out a full usage reset instead of that day's ship.
• Day 1's speedup takes the two models to roughly 50 tokens per second from about 30, and it also reaches OpenCode, Pi, Amp and Devin for people using Sign in with ChatGPT.
• The challenge lives only in Sottiaux's posts, with no stated tiers or definition of a full reset, and on October 30 the $200 Pro plan's Codex allowance drops from 20x to 10x Plus.

If you have not been following, OpenAI has spent months teaching subscribers to watch for resets. Sottiaux reset Codex limits for all paid plans in April, joking that resets cost money. In July he reset usage twice in one day after the launch that folded Codex into the ChatGPT desktop app introduced regressions in some multi-agent workflows, then again when Codex and ChatGPT Work passed 8 million active users. At DevDay on September 29, he and Sam Altman handed out a banked reset on stage.

Day 1 took GPT-6 Astra and GPT-6.1 Sol from about 30 to about 50 tokens per second

Sottiaux announced the first result of the sprint with a short line: "Let the improvements begin." In a follow-up post he said the two models now generate approximately 50 tokens per second, up from about 30. The New Stack points out that, taken literally, that is closer to a two-thirds jump than 50%, and that both figures look like rough estimates rather than benchmark results.

The speedup reaches beyond OpenAI's own apps: subscribers who connect through Sign in with ChatGPT get it in OpenCode, Pi, Amp and Devin too. The choice of model carries some irony. Two days before the challenge began, Sottiaux apologized for a slow start with GPT-6.1 Sol, blamed a major load spike in its first two days, and announced a global reset for paid ChatGPT accounts.

OpenAI will cut the $200 Pro allowance from 20x to 10x Plus on October 30

Resets are frequent enough that unofficial services track them, and one put up a 28-day scoreboard within hours of the announcement. Developers saw the gamble at once. One joked that Sottiaux had turned Codex usage into gambling and suggested burning through the allowance every day while hoping OpenAI misses a ship.

Others would rather have higher limits in the first place. One Plus subscriber said the five-hour cap already dictates when they can work, and Mark Kretschmann, a software engineer at ASYS Group, read the pledge as 28 Codex resets in 28 days.

The timing is awkward. OpenAI said last week that from October 30, near the end of the 28-day window, the Codex and ChatGPT Work allowance on the $200 Pro plan drops from 20 times the Plus allowance to 10 times. It paired that with a new $500 plan offering 25 times Plus and an Ultrafast tier for GPT-6 Astra promising up to eight times the standard speed in Codex.

Theo Browne says Opus became the first model to pass 50% of T3 Code traffic

Not everyone is rooting for the reset button. The pseudonymous X account Token Gremlin argued that 28 smaller improvements will not count for much unless OpenAI ships a model that can beat Claude Opus 5.5.

That bar has numbers behind it. T3 Code creator Theo Browne said on October 3 that Anthropic's newest Opus model had become the first to break 50% of traffic in T3 Code, with "literally half of all prompts" going to Opus.

New models do sit on Sottiaux's own list. Just before announcing the challenge, he said the team was "locking in" on simplification, greater efficiency that translates into more usage, "groundbreaking features" and new models, adding that "feedback is clear that you all want things to get simpler."

Codex runs shorter rolling windows alongside weekly limits

Codex meters usage on two clocks. A shorter rolling window caps how much you can do over a stretch of hours, while a weekly limit caps the total. Think of a gym that allows one session a day and five a week: getting today's session back does nothing if you have already used up the week.

That gap has tripped users before. After the April reset for "ALL paid plans," one Plus user filed a GitHub issue saying their weekly limit stayed on its normal cycle and Codex offered no way to tell whether the special reset had applied. According to the issue, OpenAI Support later clarified that the reset did not cover every quota type for every user.

The rules of this bet are thin. The challenge sits outside OpenAI's terms, one tracker set the October 5 to November 1 dates on its own call, and Sottiaux has not named the tiers or said what a full reset restores. In our view, a reset that names no quota type is the weak spot of the design, since the April reset for all paid plans did not cover every quota type for every user.

Day 2 and the October 30 cut

The next checkpoint is Day 2, which needs another clear improvement relevant to most Codex and ChatGPT Work users, or the reset comes due. The later one is October 30, when the Pro allowance drops to 10 times Plus, near the end of the 28-day window. Sottiaux has listed new models among the team's focus areas, but no release date has been given for one.