OpenAI is bringing back a rolling five-hour usage limit for ChatGPT Plus subscribers using Codex and ChatGPT Work. The change reverses a temporary July period in which Plus users were governed primarily by a weekly allowance, and it means heavy users must now track both a short-term window and their broader weekly quota. 101824
What changed on August 25
The five-hour cap now applies again to ChatGPT Plus usage across Codex and ChatGPT Work. It is an additional restriction, not a replacement for the weekly allowance. 182224
This change is narrower than a general ChatGPT throttle: ordinary ChatGPT conversations are not included in the reinstated Work and Codex restriction. The affected products are the more compute-intensive experiences used for longer projects and coding workflows. 22
How the five-hour and weekly limits overlap
Plus users now have two separate gates:
- Rolling five-hour limit: controls usage within a short moving window.
- Weekly allowance: controls the user’s total included usage for the broader weekly period.
Hitting either threshold can stop included access. If the five-hour window is exhausted first, the user must wait for that window to reset, even if weekly capacity remains. If the weekly allowance runs out first, waiting for the five-hour window alone will not restore included usage; the weekly allowance must reset. 424
Users can check the relevant allowance and reset information in their usage dashboard or settings. Codex users can also enter /status in an active Codex CLI session to view usage information. 8
After included limits are reached, OpenAI also supports purchasing credits for eligible features. Credits are a pay-as-you-go add-on rather than a change to the underlying subscription, and included plan usage is used first. 7
Why OpenAI restored the cap
OpenAI’s stated rationale is capacity management. The company says a short-window limit helps smooth demand on its computing resources while allowing it to maintain relatively generous weekly usage allowances. 182930
OpenAI also pointed to a user-experience problem: newer or more casual users may consume an entire week’s allowance during a concentrated burst without realizing how quickly they are using it. The five-hour gate is therefore intended not only to manage demand, but also to reduce the chance that a single session unexpectedly exhausts a user’s weekly budget. 1829
What happened during the cap-free period
On July 12, OpenAI temporarily removed the five-hour restriction for eligible Codex and ChatGPT Work users, leaving the weekly quota as the main constraint during that period. 102531
Reports from that period also described occasional early weekly-limit resets around usage milestones, rather than requiring users to wait for the normal weekly reset. Those resets were part of the temporary usage experience and should not be treated as a guaranteed feature of the restored policy. 4
How Pro plans differ
The $100 and $200 ChatGPT Pro subscriptions were not scheduled to regain the five-hour gate immediately. Available reporting says both plans would remain on weekly-only allowances for the coming months. 171826
That exemption is time-limited in the wording available here, not a permanent promise that Pro users will never receive a five-hour cap. OpenAI may revisit the policy later. 17
Enterprise and Edu usage should not be assumed to follow the Plus rules. Those plans have separate administrative and credit-based usage-management systems, including controls for limits and overages. 356
The practical takeaway for Plus users
For long Codex or ChatGPT Work sessions, the key change is that a large remaining weekly balance no longer guarantees uninterrupted access. A user can still be blocked by the rolling five-hour window, while a user who has exhausted the weekly quota must wait for the weekly reset or use available credits.
The result is a two-level allowance system: the five-hour cap spaces out intensive work, while the weekly cap limits total included usage. Pro subscribers currently avoid the short-window restriction, but that distinction is described as temporary rather than permanent. 4717