If you pay for ChatGPT Plus, Pro, or Business and you also use Codex for coding work, a change is coming that affects how your usage is counted. Starting in December 2026, your plan will no longer carry two separate budgets, one for chat and one for coding. Instead, ChatGPT and Codex will draw from one shared token allowance. This article explains what that means in plain language, what counts toward the shared pool, what stays outside it, and what you can do when you reach the limit.
One important point first: OpenAI has not announced an exact start day in December. The change is described only as starting in December 2026. It has also not published the size of each plan’s allowance in tokens, and it has not published a reset schedule for the new shared pool. Everything in this guide sticks to what OpenAI has actually said, so you will not find invented figures here. Where a number is missing, this article says so plainly, because guessing would only mislead you.
What exactly is changing
Right now, a paying subscriber effectively works with two budgets. ChatGPT conversations draw on one side, and Codex coding work draws on the other. From December, both sides draw on the same token pool. A token, in OpenAI’s own words, is the unit of information a model processes and generates. So every chat reply and every coding task consumes tokens, and from December both kinds of work spend from the same balance.
This matters most for people who use both products heavily. A long afternoon of coding with Codex and a long evening of chatting in ChatGPT will now compete for the same allowance. If you only use one of the two, the change will feel smaller, but it still changes how your usage is measured, because the two products will no longer be counted separately.
The change applies to Plus, Pro, and Business plans. OpenAI has not named Free or Go plans in this December change, so this article does not claim anything about those plans. If you are on a Free or Go plan, do not assume this change applies to you until OpenAI says so.
Before December vs from December
The table below sums up the change in one view. It is based only on what OpenAI has published through its Help Center notice, as reported at the end of October 2026.
| What | Before December | From December |
|---|---|---|
| ChatGPT usage | Counted under its own budget | Draws from the shared token pool |
| Codex coding work | Counted under its own budget | Draws from the same shared token pool |
| Instant chat | Not counted against your usage | Still not counted; it does not draw from your usage |
| How usage is measured | Separate budgets for chat and coding | One token allowance shared across ChatGPT and Codex |
| Where to check | Separate usage views | Combined remaining usage and reset times in Settings, under Usage |
| What happens at the limit | Separate limits could bite separately | One shared limit; wait for the reset, upgrade, or buy credits for eligible features |
The key row is the last one. Because the pool is shared, hitting the limit in one product now means you have hit it for both. That is the practical difference you need to plan around.
What counts and what does not
Almost everything you do in ChatGPT and Codex counts toward the shared allowance from December. Ordinary chat conversations count. Coding tasks in Codex count, whether you run Codex inside ChatGPT, in a code editor, or in the terminal. Related agent-style work that already shares Codex’s allowance, such as ChatGPT Work and similar premium features, is part of the same picture.
How much each action uses depends on two things: the model you choose and the task you ask it to do. OpenAI says more advanced models and more complex tasks generally use more of your allowance. A short, simple question costs less than a long, complex job. A long-running coding task can use substantially more than a quick request. This is worth remembering, because it means your allowance is not spent at a fixed speed. Heavy days burn faster than light days.
There is one named exception, and it is important: instant chat will not draw from your usage. OpenAI states this plainly. So quick, lightweight questions sent through instant chat do not eat into the shared pool. This is the one clear carve-out in the announcement, and it is the reason the advice later in this article suggests using instant chat for quick questions when you are running low.
What OpenAI has not done is publish a list of exact token costs per model or per task. You will not find a table of token figures in this article, because no official one exists for the December change. The honest way to learn your own burn rate is to watch your own usage, which brings us to the next section.
How to check your usage
From December, OpenAI says combined remaining usage and reset times will appear in Settings, under Usage. That single screen becomes the place to check both products at once. Here is a simple routine you can start now, before December arrives, so the new screen feels familiar when it lands:
- Open ChatGPT and go to Settings.
- Open the Usage section and look at your remaining usage and your reset time.
- Check it once a week, on the same day, so you can see how fast a normal week uses your allowance.
- In December, check the same screen again. It should now show combined usage for ChatGPT and Codex, with one reset clock, instead of two separate views.
- If you use Codex heavily, note how a coding-heavy week compares with a chat-heavy week. That comparison is your real capacity plan, because no official per-plan token figure exists.
This weekly habit is the single most useful preparation step. Without published allowance sizes, your own measured usage is the only reliable input for planning. A subscriber who knows that a heavy coding week uses roughly twice as much as a light week can plan around December with confidence. A subscriber who never looks at the Usage screen will be surprised.
What happens when you reach the limit
OpenAI documents three options when you reach the limit, and only three. Knowing them in advance means you will not panic when the limit message appears:
- Wait for the reset. Your allowance refreshes on its reset clock, and you can continue once it does.
- Upgrade your plan. A higher plan carries a larger allowance.
- Buy credits to continue using eligible features. Credits let you keep working on features that accept them, even after the allowance runs out.
Note the careful wording on the third option: credits continue eligible features. Not every feature may accept credits, so do not assume credits unlock everything. If a feature you need is not eligible, waiting for the reset or upgrading may be your only routes.
There is one more thing OpenAI says that deserves careful reading. The company says it expects most users will not need to wait for their usage to reset more often than they do now. That is a forecast, not a guarantee. Heavy users of both products at once are exactly the group most likely to notice the change, because two budgets are becoming one. Treat the expectation as reassurance for average use, not as a promise for heavy use.
How to stretch your shared allowance
Because the pool is shared, small habits now pay off twice: they save tokens in both products. None of these tips require any technical knowledge, and none of them depend on unpublished figures.
First, use instant chat for quick questions. Since instant chat does not draw from your usage, short factual questions and quick checks are better placed there than in a full chat session when your balance is low. Save your allowance for work that genuinely needs a full model reply.
Second, match the model to the job. More advanced models generally use more of your allowance, so a simple question does not need the heaviest model. When a lighter model can answer well, use it. When the task is complex, use the stronger model and accept the higher cost. This is a judgment call you will get better at as you watch your Usage screen.
Third, avoid repeating long jobs blindly. If a coding task runs long, check what it produced before asking for a fresh run. Re-running a long task from scratch spends the allowance twice. Reviewing the first result and asking for a small fix usually costs far less than starting over.
Fourth, batch your work. If you know a coding session will be heavy, do your chatting before or after it with the shared pool in mind, rather than treating the two as separate budgets. The old habit of thinking “chat is separate, so it does not matter” stops being true in December.
Fifth, keep the weekly check. The Usage screen is your early warning system. If you see the balance falling fast early in the cycle, you can switch to instant chat for quick questions and lighter models for simple work before you hit the wall. If you only check after the limit message appears, your options shrink to the three listed above.
What this change is not
It is easy to misread a pooling change as something else, so let us be clear about what it is not. This is not a price rise. No new price was announced in these reports. It is not a cut to any plan’s price or a new fee. It is a change in how usage is counted: two budgets become one shared pool.
It is also not about model pricing. The prices developers pay for API models are a different subject, and they are not part of this change. It is not about advertising either. ChatGPT advertising is a separate topic, and nothing in this announcement concerns ads. Keeping these subjects apart matters, because mixing them up leads to wrong conclusions, such as assuming your bill will rise when only the counting method changes.
It is also not a change for every plan. As stated above, OpenAI named Plus, Pro, and Business. Free and Go plans are not named in the December change, and this article makes no claim about them.
A quick word on the October background
This December change did not arrive out of nowhere. On 29 October 2026, OpenAI doubled the Work and Codex allowances included in Pro 100 and Pro 500 plans, and it cancelled a planned reduction for Pro 200 subscribers that had been announced in an email on 29 September 2026. In other words, Pro subscribers went into the December announcement with larger Codex allowances than they had been expecting, not smaller ones.
That background is worth knowing for one reason: it shows the December pooling change is happening alongside allowance increases for some Pro plans, not alongside cuts. It does not change the core fact of pooling, and it does not give us token figures for the shared pool, but it does correct any impression that OpenAI has only been tightening limits lately.
Proof
The December change was published in OpenAI’s Help Center article “Understanding ChatGPT and Codex usage limits”, updated 31 October 2026; the direct Help Center link could not be verified at writing time, so this paragraph links the report that documented the notice instead: Unite.AI reported the Help Center notice on 31 October 2026, confirming that Plus, Pro, and Business plans move to one shared token allowance across ChatGPT and Codex from December, with instant chat excluded and combined usage shown in Settings under Usage.
FAQs
Will my ChatGPT subscription cost more from December?
No new price was announced in the reports of this change. The change pools your existing allowance across ChatGPT and Codex; it does not add a fee. If you choose to upgrade your plan or buy credits at the limit, those are your choices, not a forced price rise.
Does this change apply to Free or Go plans?
OpenAI named only Plus, Pro, and Business plans in the December change. Free and Go plans were not named, so do not assume the change applies to them. Wait for OpenAI to say so before acting on it.
What is the exact start date in December?
OpenAI has published no exact start day. The change is described only as starting in December 2026. Watch the Usage screen in Settings during the first weeks of December to see when the combined view appears for your account.
What should I do if I hit the shared limit?
You have three documented options: wait for the reset, upgrade your plan, or buy credits to continue eligible features. Before that point, check Settings, under Usage, weekly, use instant chat for quick questions, and match the model to the job so your allowance lasts longer.