OpenAI's Codex lead pledged to ship one improvement a day for 28 days, or hand out a free usage reset

Running out of weekly Codex usage before the week is over now carries a public promise from the team that sets the limits.

Thibault Sottiaux, head of product and platform for Codex at OpenAI, posted on X on 4 October that his team will "either ship one thing that is a clear improvement and relevant for most codex/work users or ship a full reset" every day for the next 28 days. He closed the post with "Let the improvements begin." The pledge is on Sottiaux's personal account rather than any OpenAI page, and the company has given the window no official name.

Day one is a speed change on the two main models

Sottiaux said on day one that the team had made GPT-6 Astra and GPT-6.1 Sol "about 50% faster by default" inside Codex. A follow-up post the same day gave the number: the two models now generate "50 TPS instead of 30 TPS," his shorthand for tokens per second. The figures do not match each other, since 30 to 50 is a 67 percent increase rather than a 50 percent one, and Sottiaux presented both as rough estimates rather than benchmark results.

Developers who sign in to third-party coding tools with "Sign in with ChatGPT" get the same speed change. The New Stack names OpenCode, Pi, Amp, and Devin.

The 28 days end two days after the $200 Pro plan drops to half its old allowance

OpenAI said on 29 September that from 30 October it will cut the Codex and ChatGPT Work allowance on the $200 Pro plan from twenty times the Plus allowance to ten times. Pro subscribers also lose half of their weekly GPT-6 Pro chat allowance, going from 200 messages to 100. Existing subscribers keep their current limits until 29 October and then receive 62,500 one-time usage credits, which OpenAI values at $2,500 and expires on 31 December. The same announcement introduced a $500 plan at 25 times the Plus allowance and an Ultrafast tier for GPT-6 Astra that OpenAI says runs up to eight times the standard speed in Codex.

Several developers have described the sprint as compensation for that cut. Mark Kretschmann, a software engineer at ASYS Group, posted on X that he treats the pledge as "28 Codex resets in 28 days." A tracker set up within hours of Sottiaux's post has fixed its own window of 5 October through 1 November, and says it is picking those dates because OpenAI has not.

What "a full reset" covers is not settled

Sottiaux has not said which plans would get a reset. OpenAI has not said either. The question matters because Codex runs more than one usage window at a time: a weekly limit, and shorter rolling five-hour windows on some plans. After OpenAI's April reset for "ALL paid plans," one Plus user filed a GitHub issue because their weekly limit had not moved, and OpenAI Support later said the reset had not covered every quota type for every user.

Some developers want a new model more than a reset. The pseudonymous X account Token Gremlin said that 28 small improvements would not count for much "unless OpenAI ships a model that can beat Claude Opus 5.5." T3 Code creator Theo Browne posted on 3 October that Anthropic's Opus 5.5 had become the first model to pass fifty percent of traffic in his tool, with "literally half of all prompts" going to Opus.

Source

Primary source: Tibo Sottiaux on X, 4 October 2026, with the day-one speed post at Tibo Sottiaux on X, 5 October 2026. Reporting: Developers are secretly hoping OpenAI fails to ship this month, Amanda Caswell, The New Stack, 5 October 2026. Background on the pricing change: OpenAI halves $200 plan allowance, launches $500 plan, The New Stack, 29 September 2026.