OpenAI has brought back the five-hour usage limit for ChatGPT Plus accounts on two of its heavier surfaces — ChatGPT Work and Codex — with the change taking effect today, 25 August. For roughly the past few weeks Plus subscribers on those two products had been running against only a weekly usage cap, with no shorter rolling window sitting on top of it. That shorter window is now back.
What the ChatGPT Plus five-hour limit covers on Codex and ChatGPT Work
This is narrower than the shorthand doing the rounds suggests. The announcement names ChatGPT Work and Codex specifically. If your Plus subscription is mostly a chat box you type questions into, this particular change is not the thing that will stop you mid-sentence — the five-hour window applies to the two compute-heavy surfaces built for longer, tool-using sessions.
The two limits now stack. A Plus user hits whichever arrives first: the rolling five-hour allowance, or the weekly cap that has been in place throughout. Once either is exhausted, 9to5Mac reports that the options are to wait for the next reset cycle or buy additional credits, with OpenAI also resetting limits for free on occasion and banked resets available through some promotions and referral offers.
The announcement came from Thibault “Tibo” Sottiaux, OpenAI’s engineering lead for Codex and ChatGPT, in a post on X. His stated reason:
This is necessary as (a) the 5h limit allows us to smoothen the load on our compute, allowing to keep the plan generous in terms of weekly usage and (b) users on the Plus plan are relatively casual and new users, but then also just accidentally eat through their whole weeks usage and then are confused, making it not a great experience.
Who is exempt
Sottiaux’s post enumerates three tiers. Plus gets the five-hour limit back. The Pro $100 and Pro $200 subscriptions keep the five-hour limit not enabled for the upcoming months. Enterprise and Edu accounts were not part of the post at all; 9to5Mac notes those run on a separate credit-based usage system.
It is worth being plain about how that reads from the paying side. Part (a) of the explanation is a straightforward compute-cost argument, and it is at least stated honestly rather than dressed up as a feature. Part (b) is less convincing: framing a tightened cap as protection for “casual and new users” who might otherwise burn through their week is a customer-experience justification for something that, in practice, reduces what a fixed monthly subscription delivers mid-cycle. The tiers that keep the looser behaviour are the $100 and $200 Pro subscriptions. Whatever the engineering rationale, the outcome is that the cheapest paid tier absorbs the throttling first.
What Plus subscribers should do
If you use Codex or ChatGPT Work for long sessions, the practical change is pacing rather than total volume. Front-loading a big refactor or a multi-hour research task into one uninterrupted block is now more likely to hit a wall than the same work spread across the day, because a five-hour window resets far sooner than a weekly one. Anyone who has structured a workflow around the unlimited-within-the-week behaviour of the last few weeks will want to check where their usage stands before starting something long.
OpenAI has been adjusting these ceilings repeatedly — Sottiaux notes in the same post that he had flagged this return earlier and then postponed it. For a subscription product, that is the underlying problem: the price is fixed and published, but the amount of work it buys keeps moving, and this time it moved on the entry paid tier while the more expensive ones were left alone. If you have been treating a Plus subscription as a predictable monthly capacity, this is a reminder that it is not one.
Elsewhere in ChatGPT news, we recently covered ChatGPT Computer History, the opt-in macOS activity-recording feature.
Source: @thsottiaux on X, via 9to5Mac
