Estimated read time: 7 minutes
If you run your business on a twenty dollar ChatGPT subscription, your Monday just got a little smaller. On August 24, OpenAI’s engineering lead for Codex and ChatGPT, Thibault Sottiaux, announced on X that the five-hour usage limit was coming back for Plus subscribers on both Codex and ChatGPT Work, effective the next day. It went live August 25.
This is not a price increase. Nothing on your invoice changed. What changed is the shape of what you are allowed to use, and for a lot of small operators that turns out to matter more than the number on the bill.
Table of Contents
What actually changed
For several weeks over the summer, Plus subscribers using Codex and ChatGPT Work operated under a single ceiling: a weekly usage quota. Burn through it slowly or burn through it in one afternoon, the platform did not particularly care. That flexibility is gone.
As of August 25, Plus accounts face two ceilings at once. There is still the weekly quota. Layered on top of it is a rolling five-hour window with its own limit. Hit either one and you stop, full stop. Your options are to wait for the relevant cycle to reset, or buy additional credits.
The five-hour cap does not replace the weekly quota. It sits underneath it as a rate limiter. Think of it less like a smaller gas tank and more like a governor on the engine: the tank holds the same amount, you just cannot floor it.
OpenAI has occasionally reset user limits for free, and it has handed out banked resets through promotions and referral offers. During the window when the five-hour cap was suspended, the company also reset weekly usage early a few times to mark milestones, including passing another million active users across Codex and ChatGPT Work after the two platforms were unified. Those are gestures, not guarantees. You cannot build a workflow on the hope of a surprise reset.
Why OpenAI says it did this
Sottiaux gave two reasons, and they are worth taking at face value because they explain the shape of what comes next.
The first is infrastructure. In his words, the five-hour limit lets OpenAI “smoothen the load on our compute,” which in turn lets the company keep the weekly allowance generous. That is a straightforward capacity argument. A platform serving hundreds of millions of people cannot let a small percentage of heavy users spike demand into a wall. Rate limiting is how every utility on earth manages peak load, and inference is starting to behave a lot like a utility.
The second reason is stranger, and more revealing. Sottiaux described Plus users as relatively casual and relatively new, and said they sometimes accidentally consume an entire week’s usage in one sitting and then find themselves confused. The cap, in this framing, is partly a guardrail against your own enthusiasm.
That is a defensible product decision. It is also a quiet admission about who OpenAI thinks is on the Plus tier. If you are a solo consultant running client work through ChatGPT Work eight hours a day, you are not the user this policy was designed around. You are the edge case it was designed to contain.
Who is affected and who is not
The restored cap applies to Plus subscribers on Codex and ChatGPT Work. That is the twenty dollar tier, the one most freelancers, solopreneurs, and small teams actually pay for.
The Pro plans at one hundred and two hundred dollars a month are exempt, and OpenAI has said the five-hour limit will stay off for those tiers for the upcoming months. That is explicitly a temporary posture, not a permanent promise. Read it as “not yet” rather than “never.”
Sottiaux did not mention Enterprise or Edu accounts, which run on a separate credit-based usage system with its own mechanics for requesting higher limits. If your business is on one of those plans, this announcement does not change anything for you today.
So the affected group is narrow and specific: people paying the smallest amount of money for the highest ratio of work per dollar. Which is to say, exactly the small business owners who found the best value in the product.
Why a rolling cap hurts small operators differently
A weekly quota is a budget. A five-hour rolling cap is a schedule. Those are very different constraints, and the difference lands hardest on people whose work is lumpy.
Large companies smooth their AI usage across dozens of seats and a predictable calendar. Nobody at a two thousand person firm is doing all of the month’s analysis on the third Tuesday. Small operators do exactly that. You land a client, you have four days, and you compress a month of thinking into a weekend. That is not bad planning. That is what the business looks like.
A rolling window punishes compression specifically. The same total volume of work, spread evenly, sails through untouched. Bunched up, it hits a wall at hour three of your best working day. The tool does not care that the deadline is real.
There is a second-order cost that is harder to see on a spreadsheet: the mental overhead of rationing. Once you know there is a meter running, you start editing yourself before you type. You skip the exploratory question. You do not run the second draft. You save your good prompts for when they count. That is a tax on the exact behavior that made the tool valuable in the first place, and it does not show up in any usage report.
The real lesson: you are renting compute, not owning it
The specific policy matters less than the pattern behind it, and the pattern is now unmistakable across the industry.
Terms on AI subscriptions are not stable. They are provisional. Limits get suspended when a company wants growth, restored when it needs capacity, and adjusted again when the economics move. In the past year the sector has walked through price increases, tier restructuring, model deprecations, and rate limits appearing and disappearing on roughly a monthly rhythm. This latest change was announced one day before it took effect, in a post on social media, by an engineer.
That is not a criticism of OpenAI in particular. It is the nature of a market where the underlying cost of serving a request is enormous, volatile, and only loosely connected to what customers are being charged. Somebody is absorbing that gap, and the terms of absorption keep getting renegotiated.
The practical implication for a small business is unglamorous but important. Treat any AI subscription the way you would treat a contractor with no contract: useful, currently affordable, and not something to build a load-bearing wall on. If a single vendor’s rate limit can stop your delivery, that is not a vendor problem. That is a design problem in your operation.
This is the same reasoning behind auditing which of your SaaS tools are exposed to AI disruption. The question is never whether a tool is good. It is what happens to you the week it changes.
Five moves worth making this week
None of these require a budget increase. All of them reduce how much a policy change can cost you.
1. Find out what your actual usage looks like. Most people have no idea whether they are near a limit or nowhere close. Before you react, spend a week noticing when you hit walls and when you do not. A large share of small business users will never touch the five-hour cap, and panic-upgrading is the most expensive possible response to a limit you would not have reached.
2. Move your heaviest work out of one window. If you routinely do four hours of dense generation in a block, split it. Draft in the morning, refine after lunch. The rolling window resets continuously, so spacing work out is often enough to make the cap invisible.
3. Keep a second model available. Not as a full replacement, as a pressure valve. Claude, Gemini, and several open-weight options run well enough for drafting, summarizing, and rewriting that hitting a wall on one platform should be an inconvenience rather than a stoppage. Our roundup of AI writing tools for small business covers where the practical differences actually are.
4. Do the credit math before you buy credits. Additional credits are the friction-free option, which is exactly why you should price them against a Pro plan and against a competing subscription before making it a habit. Convenience purchases compound quietly.
5. Write down which parts of your business would stop. Literally list the deliverables that cannot ship if your primary AI tool is unavailable for six hours. If that list has client-facing items on it, you have found this month’s most valuable process improvement, and it has nothing to do with which model you use.
What to watch next
Two things.
First, whether the five-hour cap eventually reaches the Pro tiers. OpenAI framed the exemption as lasting for the upcoming months, which is a conspicuously soft commitment. If capacity pressure continues, the reasoning that justified capping Plus applies just as well one tier up.
Second, whether competitors follow. Rate limiting is contagious. When one major platform establishes that a rolling window is normal, it becomes much easier for the others to introduce their own without absorbing the full public relations cost. Watch the next two quarters of pricing pages more carefully than the next round of model announcements.
The model capabilities will keep improving. The terms under which you are allowed to use them are the variable that actually touches your business, and it is the one almost nobody is tracking.
Frequently asked questions
Does the five-hour limit replace the weekly quota?
No. Both apply at the same time. You can hit either one independently, and hitting either one stops you until that specific cycle resets or you buy additional credits.
Which plans are affected?
Plus subscribers using Codex and ChatGPT Work. The one hundred and two hundred dollar Pro plans are exempt for at least the next several months. Enterprise and Edu accounts run on a separate credit-based system and were not mentioned in the announcement.
When did this take effect?
The change was announced on August 24, 2026, and took effect on August 25, 2026.
What happens when I hit the limit?
You wait for the cycle to reset or purchase additional credits. OpenAI has also reset limits for free on occasion and offered banked resets through promotions and referrals, but neither is something to count on.
Should I upgrade to Pro because of this?
Probably not as a reflex. Track your real usage for a week first. Many Plus users on typical small business workloads will not reach the five-hour cap, and the upgrade is a fivefold price increase to solve a problem you may not have.
Is this a sign that AI subscription prices are going up?
It is a sign that AI subscription terms are unstable, which is a broader and more useful thing to know. Capacity constraints get expressed as price changes, tier changes, or usage limits depending on what a company thinks the market will tolerate. Plan for all three.
Related Coverage
- Which of Your SaaS Tools Are About to Get Eaten by AI: the same underlying question, applied to your whole software stack.
- Best AI Writing Tools for Small Business: what to keep in reserve when your primary tool hits a wall.
- AI Agents Explained: The 2026 Rundown: why compute demand is climbing in the first place.
Faceted Media Magazine covers business, AI, and entrepreneurship for the people building what’s next.