Hajra Naz
OpenAI restored a five-hour usage limit on Codex and ChatGPT Work for ChatGPT Plus subscribers on August 25, 2026, ending several weeks in which the weekly quota was the only ceiling on those two products. Thibault “Tibo” Sottiaux, OpenAI’s engineering lead for Codex and ChatGPT, announced the change on X the night before it took effect. The $100 and $200 Pro tiers stay exempt for the coming months.
What changed for Plus subscribers
OpenAI temporarily lifted the five-hour restriction on July 12, leaving the weekly quota as the main constraint for eligible Codex and ChatGPT Work users. During that window the company also reset weekly usage early on several occasions to mark milestones, including passing another million active users across the two now-unified platforms.
From today, Plus accounts hit whichever ceiling comes first. Once a user reaches either the five-hour or the weekly limit, the options are to wait for the next reset cycle or buy additional credits. OpenAI occasionally resets limits for free, and banked resets can be earned through certain promotions and referral offers, according to 9to5Mac.
Sottiaux did not mention any change for Enterprise or Edu accounts, which run on a separate credit-based usage system.
OpenAI’s stated reasoning
Sottiaux gave two reasons. The first is infrastructure: the five-hour limit lets OpenAI smooth the load on its compute, which in turn lets it keep the weekly allowance generous. The second is user experience. He described Plus subscribers as relatively casual and new users who can accidentally burn through a full week of usage in one sitting and then find themselves confused about why the product stopped working.
Both explanations point at the same underlying pressure. Agentic workloads consume far more compute per session than ordinary chat, and Codex sessions in particular can run long. Rationing by the hour spreads that demand out instead of letting a small number of heavy users concentrate it.
Why an hourly ceiling lands differently on offshore delivery work
For a freelancer in Karachi or Dubai billing a client in London or Toronto, the difficulty was rarely about price. A lot of offshore delivery happens in long uninterrupted blocks, often at odd hours, because that is when the client’s feedback arrives and when the revision window is open. A cap that resets every five hours cuts across exactly that pattern.
The obvious escape route is expensive. Plus costs $20 a month, while the tiers that keep the five-hour limit switched off start at $100. For a mid-career freelancer in Pakistan, that gap is a meaningful share of monthly earnings, not a rounding error. The realistic response is a workflow change rather than a plan upgrade: split heavy generation across separate sessions, assemble prompts and reference material before starting the clock, and reserve the expensive multi-step agent runs for the parts of a job that genuinely need them.
The timing is notable given that OpenAI is simultaneously pushing AI agents toward everyday office work. The company is asking people to hand more of their workflow to agents while tightening how much agent time the entry-level paid plan buys.
What to watch before the end of the year
Sottiaux framed the Pro exemption as holding “for the upcoming months”, which is a commitment with an expiry built into it. If compute demand keeps climbing as agents move from developer tooling into general knowledge work, the question worth tracking is whether Pro stays exempt or whether a five-hour window becomes the default across every consumer tier. Anyone building a client-facing process on top of Codex should design for the ceiling rather than assume the exception will hold.
