If you run Codex on a ChatGPT Plus plan, your sessions are about to hit a wall again after weeks without one.

Tibo Sottiaux, OpenAI’s engineering lead for ChatGPT and Codex, posted on X that a five-hour usage cap is returning for Plus subscribers. It applies across Codex and ChatGPT Work, OpenAI’s work-focused ChatGPT tier that shares the same usage pool. The cap throttles how much you can use within any rolling five-hour window, separate from your normal weekly allowance.

What’s actually changing

Plus subscribers currently only run into one limit: a weekly usage allowance that resets on a fixed schedule. Once the five-hour cap returns, Plus users will hit two ceilings instead of one. You can burn through your five-hour allotment well before your week’s allowance runs out, and then have to wait out that shorter window before continuing.

Pro subscribers, on OpenAI’s $100 and $200 monthly tiers, are staying exempt from the five-hour cap “for the upcoming months,” according to Sottiaux. Their usage is still governed by the weekly allowance only.

Sottiaux gave two reasons for bringing the cap back. First, spreading usage across five-hour windows helps OpenAI manage how much server capacity everyone’s requests draw on at once, without having to shrink the weekly allowance for everyone. Second, he said Plus users skew toward newer, more casual accounts that sometimes burn through an entire week’s allowance in one sitting, then get confused when they’re locked out for days. The five-hour cap is meant to catch that earlier, in a smaller, faster-resetting window.

9to5Mac independently confirmed the change, reporting that OpenAI had not yet published it as a formal blog post or help center update at the time of writing. Sottiaux’s post is currently the only place OpenAI has stated the change and its reasoning.

This is a reversal, not a new policy

BuilderWithin covered this same cap in July, when OpenAI suspended it after Codex and ChatGPT Work crossed 9 million active users. That suspension ran for roughly six weeks, alongside repeated weekly-allowance resets while OpenAI fixed bugs that made its GPT-5.6 Sol model burn through usage faster than intended.

Sottiaux’s post frames this as bringing the cap “back,” not introducing something new. If you started using Codex during that suspension, this will be the first time you’ve run into the five-hour wall at all.

What builders should do next

Don’t guess at how much the five-hour cap will actually cost you. Pick one bounded task you’d normally hand to Codex, something like migrating the tests for a single API endpoint, and run it once the cap is live. Track two things: how much of the five-hour window it consumes, and how much of your weekly allowance is left afterward.

If that one task already eats a large share of both, that’s a concrete signal your workload doesn’t fit comfortably inside Plus’s cap, not a guess. Compare that against OpenAI’s Codex rate card, which lists the exact quotas per plan, before deciding whether a Pro plan’s higher price is worth avoiding the wait.

Where this fits

Every coding agent vendor is running the same experiment right now: how much usage can they give away before compute cost or account abuse forces a pullback. Anthropic just went the other direction, extending Claude Code’s larger limits through the end of August. OpenAI is doing the opposite for its cheapest paid tier. Neither move tells you what your subscription will cost or deliver next month. Track your own usage against the task you actually need done, not the announcement of the week.

End of article