 |
| AI-generated illustration of a laptop with coding on one side and chat messages on the other, both feeding into a single shared usage meter. |
If you pay for ChatGPT and you also use Codex, December brings a change that is easy to miss and easy to misunderstand. OpenAI is not announcing a new product, a new price, or a new model in this update. It is changing the way your existing paid allowance is counted. Starting in December, Plus, Pro and Business plans will share one usage allowance across ChatGPT and Codex, measured in tokens, instead of keeping chat and coding in separate buckets.
That sounds small. In practice, it changes how a working day adds up. A long coding session in the afternoon and a long run of chat questions in the evening will now draw from the same pool. If you use only one of the two products, or you use both only lightly, the change may pass unnoticed. If you split your day between writing, planning and coding with AI help, you will want to understand the new meter before December arrives, because the habits that kept you comfortably inside two separate limits may not keep you inside one shared limit.
As an Amazon Associate, we earn from qualifying purchases.
Prices and savings can change. Check the retailer for the current price.
There is also an important caution to keep in view. OpenAI has published the shape of the change, but not the numbers behind it. It has not given an exact December start date, and it has not published a token size for any plan. That gap matters. It means nobody outside OpenAI can yet tell you, with confidence, exactly how many chats or coding tasks your plan will cover after the change. This article therefore sticks to what OpenAI's Help Center wording actually says, separates that from what other coverage reports, and shows you how to measure your own usage in November so you are not guessing in December.
What is actually changing
The core sentence, as quoted in coverage of OpenAI's Help Center article "Understanding ChatGPT and Codex usage limits", is this: "Starting in December, Plus, Pro, and Business plans' usage allowance will be shared across ChatGPT and Codex as we bring the agentic capabilities of Work into the main ChatGPT experience."
In plain language, three things are being joined together. First, the plans in scope are Plus, Pro and Business. Second, the products in scope are ChatGPT and Codex. Third, the unit of measurement is the token, which OpenAI describes as the unit of information a model processes and generates. In other words, you are not moving to a fixed number of messages. You are moving to a pool of token capacity that different models and different tasks consume at different rates.
This is a consolidation, not a brand-new idea. OpenAI's Help Center already says that Work and Codex share the usage allowance included with a plan, and that ChatGPT for Excel and Workspace Agents can share that same agentic allowance where a plan includes them. What December adds is ordinary ChatGPT conversation with the advanced models, folded into that same shared pool. The exception, stated plainly by OpenAI, is Instant chat: "Instant chat will not draw from your usage." That exception is worth remembering, because it gives you a genuine, documented way to handle quick questions without touching the shared pool.
What is not changing, at least in the wording published so far, is equally important. OpenAI has not announced a price change alongside this allowance change. It has not said that Free, Go, Enterprise or Edu plans are covered by this particular passage, so do not assume the same rule applies to them. And API usage remains a separate matter: OpenAI's Work and Codex usage guidance notes that API usage is billed separately at API rates, so code run through an API key is not the same bucket as code run through a ChatGPT plan.
Why tokens, not messages, change the maths
When a limit is counted in messages, it is easy to plan: ten questions means ten messages, whatever they contain. A token allowance does not work like that. OpenAI says how much of your allowance you use "will depend on the model you choose and the task you ask it to do", and that "more advanced models and complex tasks generally use more of your allowance."
That has a practical consequence that catches people out. A short question to a lighter model and a long, multi-step coding task with a powerful model are not comparable units, even though both might feel like "one use" of the tool. A coding agent that reads a large codebase, runs tests, revises its work and produces a long answer will consume far more tokens than a one-paragraph chat reply. Similarly, asking the most capable model for a task that a lighter model could handle comfortably will burn the shared pool faster for no extra benefit.
OpenAI's own estimates make this concrete, and they also show why you should treat them as ranges, not promises. For Plus and Standard Business plans, OpenAI estimates local messages within a five-hour window at roughly 5 to 45 messages for GPT-6 Astra, 15 to 160 for GPT-6.1 Sol, and 350 to 3,000 for GPT-6 Luna. Those are not guaranteed counts. OpenAI says explicitly that they are estimates, not guaranteed message counts, and that usage varies with the task, the model and the settings. Cloud tasks may use more allowance than local messages, and weekly limits may also apply on top of the five-hour window, depending on the plan.
The lesson is not to memorise those ranges. It is to stop thinking in messages and start thinking in cost per task. Before December, ask yourself which tasks genuinely need the most capable model, which tasks are routine enough for a lighter model, and which tasks are really just quick questions that Instant chat could handle without touching the pool at all.
Who will feel it most, and who probably will not
OpenAI's own expectation, quoted in its Help Center article, is that "we expect most users won't need to wait for their usage to reset more often than they do now." Treat that sentence carefully. It is an expectation from the company making the change, not a guarantee, and it arrives without the token numbers that would let an independent reader check it. For a casual user who chats a little and rarely opens Codex, that expectation is probably reassuring and probably right. For a heavy user of both products, it is not a reason to stop measuring.
The people most likely to feel the change are developers and small teams who already live in both products in the same day. Consider a typical pattern: Codex in the morning to build or refactor a feature, ChatGPT Work in the middle of the day to turn notes into a deliverable, and ordinary ChatGPT in the evening to draft, summarise or plan. Today, the chat portion and the agent portion are metered separately. In December, that whole day draws from one pool, so a heavy coding morning leaves less room for an evening of chat work, and vice versa. The risk is not that the tool stops working. The risk is that you hit the limit at the worst moment, in the middle of a task you needed to finish that day.
Pro users have a partial cushion that is worth noting, but it is not a complete exemption. OpenAI's Work and Codex guidance says that Pro 100, Pro 200 and Pro 500 currently have no five-hour usage limit in Work and Codex, while still having an included usage allowance to check in Settings. "Currently" is doing real work in that sentence: it describes the present position, not a promise about December. Even without a five-hour window, a Pro user still has a total included allowance, and that allowance will now be shared. Do not read "no five-hour limit" as "unlimited".
At the other end, the people least likely to notice are those who use ChatGPT mainly for Instant chat, or who use Codex only occasionally. If your usage is light in both products, one shared pool may even feel simpler, because you will see a single balance instead of two separate ones. The change is not, by itself, a reason to upgrade, downgrade or cancel. It is a reason to look at your actual November usage and decide from evidence, not from headlines.
What happens when you hit the limit
OpenAI lists three options when the shared allowance runs out: wait for it to reset, upgrade your plan, or buy credits to continue using eligible features. Each of those options has a trade-off that is worth thinking through before you need it, not while a deadline is pressing.
Waiting is free, but it is only useful if you know when the reset happens. That is why OpenAI points users to Settings, then Usage, where combined remaining usage and reset times will be shown. Get into the habit of checking that screen in November, while the limits are still separate, so you learn your own pattern: which days you run low, which model choices push you there, and how long a reset actually takes for your plan. A reset time you discover during a calm week is useful information. A reset time you discover during a deadline is a delay.
Upgrading is the simplest option on paper, but it only helps if the higher plan genuinely gives you more of the allowance you actually use. Because OpenAI has not published December token sizes, you cannot yet compare two plans by number. You can, however, compare them by your own November evidence: if you hit chat limits but never touch Codex, or the reverse, a shared pool may change which plan makes sense for you. Wait for the December numbers, or at least for your first December usage screen, before paying more on the strength of a guess.
Buying credits is the middle path for an occasional heavy week. OpenAI presents credits as a way to continue using eligible features without changing plans. Coverage of OpenAI's credits pages notes that credits can be bought from Settings and Usage in ChatGPT or from Usage and Billing in the Codex app, but the exact eligibility of every feature, and the price that applies to your plan, should be checked in your own account rather than assumed from a summary. Credits are a tool for a spike, not a substitute for a plan that is routinely too small. If you find yourself buying credits every week, that is evidence that your plan, or your model choices, need to change.
One more practical point: a Codex task already in progress is not necessarily cut off mid-sentence. Coverage of OpenAI's Codex help notes that Codex can usually finish the turn it is working on, subject to fair-use limits, after which the limit banner or usage page shows your options. Do not rely on that as a safety net for long jobs. Break large coding work into smaller, checkable steps, save and commit work as you go, and avoid starting a long agent run when your usage screen already shows you close to the limit.
How to prepare in November: a practical plan
November is your measurement month. You have roughly four weeks before the change, and the most valuable thing you can do is learn your own numbers while the old, separate limits still apply. The steps below need no code and no new subscription.
First, find your usage screen now. In ChatGPT, open Settings, then Usage, and note what it shows today: your remaining allowance, your reset times, and whether your plan shows five-hour and weekly windows. Do the same for Codex, if you use it separately. Screenshot or write down the numbers once a week. After a few weeks, you will have a baseline that no article, including this one, can give you, because it is yours.
Second, separate your work into three buckets for a week. Bucket one is quick questions and short answers: these are candidates for Instant chat, which will not draw from the shared pool. Bucket two is ordinary chat work that needs an advanced model: drafting, analysis, planning and revision. Bucket three is agentic work: Codex coding tasks and ChatGPT Work deliverables that run for many steps. At the end of the week, look at which bucket actually dominates your usage. Many people discover that one long agentic session outweighs a whole week of short chats. That discovery tells you where to economise.
Third, test the lighter model deliberately. For one routine task a day, try the lighter model first and only move up if the result is not good enough. You will quickly learn which of your tasks genuinely need the most capable model and which were simply using it out of habit. This single habit is often the difference between comfortably inside the allowance and hitting the limit, because the estimated message ranges show how wide the gap between models can be.
Fourth, protect your coding work from a mid-task limit. Before a long Codex session, check Settings and Usage, commit or save your current work, and break the job into steps that can be checked one at a time. If you are close to a reset, consider doing the quick chat work first with Instant chat, or postponing the heavy coding run until after the reset, rather than starting it and hoping.
Fifth, decide in advance what you will do at the limit. Write down your own rule now: for example, "If I hit the limit, I will wait until the reset shown in Settings, unless a client deadline is the same day, in which case I will buy credits for that task only." A rule made calmly is far more useful than a decision made under pressure, and it stops a single heavy day from turning into an unplanned upgrade.
The traps to avoid
The first trap is assuming a December 1 start date. OpenAI says only "starting in December". Coverage has repeatedly noted that no exact day has been published. Do not plan a hard cutover for December 1, and do not assume that a task on December 2 will already be under the new rules, or that a task in late November is already affected. Check the Help Center article and your Settings and Usage screen for the actual rollout day when December begins.
The second trap is treating OpenAI's expectation as a guarantee. "We expect most users won't need to wait for their usage to reset more often than they do now" is a forecast about most users, not a promise to you. If you are a heavy user of both products, you are, by definition, not the average user that sentence is about. Measure, and let your own numbers override the general expectation.
The third trap is assuming the change covers every plan. The Help Center passage names Plus, Pro and Business. It does not name Free, Go, Enterprise or Edu in the passage reported. If you are on one of those plans, or you manage a team on a Business or Enterprise seat, check your own plan documentation and your admin settings rather than applying this article's summary to your account.
The fourth trap is confusing the shared ChatGPT and Codex allowance with API billing. If you also use OpenAI's API, that usage is billed separately at API rates, according to OpenAI's guidance. A shared subscription allowance does not automatically mean your API bill changes, and an API bill does not tell you how much subscription allowance you have left. Keep the two in separate columns in your own budgeting.
The fifth trap is the most human one: trying to game the meter instead of doing the work well. Splitting one sensible task into many tiny prompts, repeating failed prompts with the most powerful model, or running the same large codebase through Codex several times without checking intermediate results will all burn tokens faster than a careful, well-specified task. The cheapest token is the one you never needed to spend, and the best preparation for December is better task design in November: clear instructions, small steps, and checking results early.
What to watch for in December
When December arrives, three things deserve a check on day one. First, re-read the Help Center article itself, because OpenAI may add the exact start date and, hopefully, per-plan token figures that are missing now. Second, open Settings, then Usage, and confirm that you see a single combined balance and reset times, as OpenAI promises, rather than two separate meters. Third, run a normal working day and compare it with your November baseline. If the new meter behaves very differently from what you expected, that is the moment to reconsider your model choices or your plan, with real December evidence in hand.
It is also worth watching for clarification on two open questions. The first is whether per-model message caps in chat, such as weekly allowances for the most capable chat model on Pro, stay alongside the new token pool or disappear into it. The Help Center article reported so far does not settle that question. The second is whether the five-hour and weekly windows change in December. Again, the article does not say. Until OpenAI publishes those details, any article that claims to know them is guessing, and this article does not guess.
For teams, December is also a good moment to agree on a shared rule. If several people draw from Business seats, a shared pool makes one person's heavy coding afternoon visible in another person's evening chat work in a way that separate limits did not. A short team note — which tasks use Instant chat, which use the lighter model, and when to buy credits rather than upgrade — will prevent more friction than any individual habit.
Checklist: get ready before December
- Confirm your plan: check whether you are on Plus, Pro or Business, the three plans named in OpenAI's Help Center passage.
- Find Settings, then Usage, in ChatGPT and note your current remaining allowance and reset times.
- Record your usage once a week in November to build your own baseline before the change.
- Sort a week's work into Instant chat, ordinary advanced chat, and agentic Codex or Work tasks, and see which bucket dominates.
- Try the lighter model first for routine tasks, and only move to a more capable model when the result needs it.
- Use Instant chat for quick questions, since OpenAI says it will not draw from the shared allowance.
- Before a long Codex run, check your usage, save and commit your work, and break the job into checkable steps.
- Decide your at-the-limit rule now: wait for the reset, upgrade, or buy credits for eligible features, and in which situation you will pick each.
- Do not assume a December 1 start date, a published token size for your plan, or coverage for Free, Go, Enterprise or Edu plans.
- In December, re-read the Help Center article, check for a combined balance in Settings and Usage, and compare your first December week with your November baseline.
- Keep API billing separate in your budgeting: OpenAI says API usage is billed separately at API rates.
- If you buy credits regularly, treat that as evidence to review your plan or your model choices, not as a routine cost to ignore.
Bottom line
ChatGPT and Codex sharing one allowance in December is not a reason to panic, and it is not a reason to ignore the change either. It is a metering change that rewards the users who already work carefully: clear tasks, the right model for the job, quick questions handled by Instant chat, and long agent work broken into steps. The users who will feel it are the heavy users of both products, and the users who will barely notice are the light users and the Instant-chat users.
The honest limit of this article is the same as the honest limit of every article on this topic right now: OpenAI has published the direction, but not the December date and not the token numbers. Do not let anyone sell you certainty that OpenAI has not published. Measure your own November usage, set your own at-the-limit rule, and check Settings and Usage again when December begins. That is a small amount of preparation for a change that, handled well, should feel like a simpler single meter rather than a smaller allowance.
Reference sources
- OpenAI Help Center: Managing usage with GPT-6 Astra in Work and Codex — https://help.openai.com/en/articles/20001516-managing-usage-with-gpt-6-astra-in-work-and-codex
- 7minAI: ChatGPT and Codex Will Share One Usage Limit in December: What Counts, What Doesn't, How to Prepare — https://7minai.com/chatgpt-codex-shared-usage-limit-december/
- Skimwire: ChatGPT and Codex share one usage limit — https://www.skimwire.com/chatgpt-codex-shared-usage-allowance-december/
- Limit Digest: ChatGPT and Codex shared limit: OpenAI confirms December — https://limitdigest.com/guides/chatgpt-codex-shared-limit/
- Automater: ChatGPT and Codex Move to One Shared Token Allowance in December — https://automater.ai/news/chatgpt-codex-shared-token-allowance-december/
- CellCog: ChatGPT and Codex Will Share One Usage Limit in December — https://cellcog.ai/blog/chatgpt-codex-shared-usage-limit/
Leave a public comment