Codex has received a rare public pledge from OpenAI: on October 4, 2026, Thibault Sottiaux, OpenAI's head of products and platforms, announced on X that over the next 28 days, Codex and ChatGPT Work will ship an improvement that is clearly meaningful for most users every single day — or else carry out a full reset. Counting from the start date, the cycle runs from October 5 to November 2. A day earlier, he had laid out the team's four locked-in directions: simplify the product, raise efficiency to release more usable capacity, ship breakthrough features, and launch new models.
Why the pledge comes at this particular moment
The background is a week of accumulated user frustration. After DevDay, the Pro 200 plan's Codex and Work allowance was cut from 20 times Plus to 10 times, with eligible existing users keeping the old allowance only until October 29; the new flagship GPT-6.1 Sol also had a slow start, and Sottiaux had just issued a global reset for all paid ChatGPT accounts on October 3 as compensation. A reset can put out a fire, but it cannot rebuild trust — what users actually resent is that allowance rules became tighter and harder to predict. Writing 'a daily improvement, or a reset' as a public commitment turns the reset from an ad-hoc apology into a fallback clause: on any day the team ships nothing meaningful, it pays in quota instead.
How to judge whether the pledge holds
Its strength is verifiability: whether each of the 28 days brought an update, and whether that update is genuinely felt by most users, is something users can check against their own usage without reading a press release. The risk is equally clear: Sottiaux did not specify the reset's coverage or quota terms, so if one day's 'improvement' turns out to be a minor tweak, the pledge will be read as marketing. The coming four weeks are also dense with comparisons between Codex and Claude Code, as Opus 5.5 has sharpened the contest over capability, speed and allowances; OpenAI is answering with delivery cadence. For heavy users, the sensible move is to treat these 28 days as an observation window — track quota consumption and feature changes, a pattern already visible in how the GPT-6 family selection guide laid out the tiers — and only then decide which plan to renew, rather than being swayed by a single reset.
