Chat and coding workspaces connected to a shared glowing reservoir representing a common token budget

ChatGPT and Codex Will Share One Token Budget in December

ByAIERA Blog Editorial Team

Key Takeaways

  • The ChatGPT Codex shared token budget starts in December 2026: Plus, Pro, and Business plans will share a single usage allowance across ChatGPT and Codex, measured in tokens rather than in messages.
  • The change comes from an OpenAI Help Center article, “Understanding ChatGPT and Codex usage limits,” updated on 31 October 2026 and widely reported on 1 November — there was no launch blog post and no press release.
  • Instant chat will not draw from the shared allowance, so casual chats with Instant models should feel unchanged, while heavier models and longer, more complex tasks will use the budget faster.
  • OpenAI has published no exact start date in December and no allowance size for any plan, so the only reliable capacity planning input before December is your own measured split between chat and Codex usage.
  • When the shared limit runs out, the documented options are the same three as before: wait for the reset, upgrade your plan, or buy credits for eligible features — and combined remaining usage and reset times will appear in Settings, under Usage.

Introduction

The ChatGPT Codex shared token budget is one of those changes that sounds administrative and turns out to be structural. From December 2026, if you pay for ChatGPT Plus, Pro, or Business, your chatting and your coding will no longer draw on separate allowances. They will draw on one pool, measured in tokens, and a long afternoon of agent coding will compete directly with every question you ask in the chat window afterwards. For light users, OpenAI says, little will change. For heavy users — especially developers who live in Codex and power users who lean on the most advanced models — the way they budget their subscription is about to change.

The change is also a case study in how OpenAI now ships important product news. There was no keynote, no blog post, no press release. The announcement surfaced as an update to a Help Center support article, was spotted and reported on 1 November 2026, and has since been confirmed across multiple independent reports reading the same page. That makes it unusually important to separate what the page actually says from what commentators have inferred from it — and this article does exactly that, sentence by sentence.

ChatGPT Codex shared token budget: what OpenAI actually said

The Help Center article’s key sentence, as quoted in multiple reports, is: “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.” That sentence carries three pieces of information, and each one matters.

First, the plans: Plus, Pro, and Business only. Free, Enterprise, and Edu plans are not mentioned in the December announcement, so there is no confirmed change for them yet. If you are on Free, this change does not, on the published wording, apply to you; if you are on Enterprise or Edu, your terms are separate and unchanged by this article.

Second, the unit: tokens. Usage “will be measured in tokens, the units of information a model processes and generates.” That is a shift from counting messages or requests. A short question and a long, multi-step agent task are no longer roughly equivalent units of usage; they are different amounts of tokens, and the budget drains at different rates depending on what you do.

Third, the reason: the merge of Chat and Work. OpenAI has already announced plans to fold ChatGPT’s Work mode — the longer, multi-step, deliverable-focused mode — into the main chat interface. Work already shares an allowance with Codex, so the December change extends a sharing arrangement that already exists for Work users to all Plus, Pro, and Business chat usage. An OpenAI engineer wrote on 28 October 2026 that the team was working on the merge but that it was “not rolling out just yet” and that OpenAI would “proceed carefully.” The December token change is therefore best read as the usage-limit side of a larger interface merge, not as a standalone pricing change.

Two further sentences from the article complete the mechanics. “More advanced models and complex tasks generally use more of your allowance” — so model choice becomes a budget choice, not just a quality choice. And “Instant chat will not draw from your usage” — so the lightest, fastest chats stay outside the meter entirely. Combined remaining usage and reset times will appear in Settings, under Usage, giving subscribers one dashboard instead of two separate balances.

What has not been published — and why it matters

What the article does not say is, for planning purposes, almost as important as what it does say. OpenAI has published no exact day in December 2026 when the change takes effect. It has published no allowance size for Plus, for any Pro tier, or for Business. It has published no reset cadence beyond the promise that remaining usage and reset times will be visible in Settings. And it has not said how the new token allowances compare, in practice, with the message-based limits they replace.

That gap is why the most useful advice in the independent coverage is also the simplest: measure your own usage now. Before December, a plan effectively carries two budgets — one for chat, one for coding. From December, both draw on a single pool with one reset clock. If you know, roughly, how much of your month currently goes to Codex versus chat, you have the only before-and-after comparison that will be available until OpenAI publishes numbers or until the Settings dashboard shows your combined balance. If you do not know that split, November is the month to learn it, while the two budgets are still separate and visible.

The gap also explains the most common misreading of the change. Several commentators have described the December move as “the end of all-you-can-prompt pricing” or “the end of flat-rate ChatGPT.” That reading goes further than the article does. The article does not announce a price rise, and it does not say ChatGPT stops being a subscription service. What it says is that usage within a subscription will be measured in tokens from a shared pool, with Instant chat exempt. Whether that feels like a tighter limit or a more flexible one will depend entirely on the unpublished allowance sizes — and on how much of a given user’s work can live in Instant chat. Until those sizes are known, the careful description is “a shared token budget with unpublished sizes,” not “a price rise” and not “the end of flat-rate.”

What the current plans look like, for context

To understand who is affected, it helps to know the current plan structure as read from OpenAI’s pricing pages on 2 November 2026, in US prices before tax. Codex has no price of its own: OpenAI sells it inside ChatGPT, so what a subscriber pays for is the size of the allowance. Plus costs $20 a month and includes Codex across web, CLI, IDE extension, iOS, GitHub code review, and Slack, with a five-hour window and a weekly cap. Pro comes in tiers at $100, $200, and $500 a month, with progressively more usage and no five-hour limit on the higher tiers, and the $500 tier adds Ultrafast. Business Standard is $20 a seat billed annually ($25 monthly) with ranges similar to Plus, and Business Premium is $100 a seat billed annually ($125 monthly) with more usage.

Two recent changes to those plans are relevant context, and both should be labelled as reported rather than as confirmed on OpenAI’s own pages where the reporting is careful about that distinction. On 29 October 2026, OpenAI reportedly reversed a planned cut to Pro usage limits, with included usage on the $200 Pro tier doubling and Codex and ChatGPT Work usage doubling on the other Pro tiers, with prices unchanged. One roundup of the week notes that it has not confirmed that reversal on OpenAI pages, so subscribers should check current plan details before relying on it. The broader point stands regardless: OpenAI has been adjusting Pro allowances in the run-up to the December merge, which suggests the company is aware that a shared budget will feel tighter unless allowances are generous enough to cover both chat and coding from one pool.

When the shared limit does run out, the article documents three options, and only three: wait for the reset, upgrade the plan, or buy credits to continue using eligible features. There is no fourth option in the article, and no suggestion that the limit can be bypassed by switching between ChatGPT and Codex — because, from December, switching no longer switches pools.

Who feels this, and who does not

The impact splits fairly cleanly by usage pattern, and OpenAI’s own expectation, quoted in the coverage, is that “most users won’t need to wait for their usage to reset more often than they do now.” That is a forecast, not a guarantee, and it should be read as one. But the logic behind it is clear: a user whose work is mostly short chats, especially with Instant models, is barely touching the meter at all, and a shared pool does not change that. A user who runs long agent tasks in Codex and then asks long, complex questions in ChatGPT is the user whose two previously separate budgets now collide.

Three groups should pay particular attention. First, Plus subscribers who use Codex heavily: Plus has the smallest allowance and the five-hour window, so a shared pool is most likely to be felt there first. Second, teams on Business, where one seat’s coding and chatting both draw on that seat’s pool, and where an admin buying workspace credits becomes the pressure valve. Third, anyone who has built a workflow that assumes chat and Codex are independent — for example, doing heavy agent work in Codex while keeping ChatGPT free for questions — because that assumption stops being true in December.

For those users, the practical steps are concrete and available now. First, check Settings, then Usage, and note your current separate balances. Second, identify which of your regular tasks can live in Instant chat, since those chats are exempt and therefore represent free capacity after the merge. Third, for tasks that cannot, estimate their token cost by watching how quickly your current allowances drain on heavy days, and decide whether your current tier still fits a combined pool — or whether the December change is the moment to upgrade, to buy credits, or to move some work to the API, where pricing is per token at API rates and is separate from the subscription pool. Note, however, that Pro and Business API credits cannot pay for Codex, according to pricing-page readings, so the API is not a back door into a larger Codex allowance.

The bigger shift: from messages to tokens

Step back, and the December change is part of a wider 2026 shift in how AI subscriptions are measured. Counting messages made sense when a chat was a chat: one question, one answer, roughly comparable effort each time. It makes much less sense when a single “request” can be a multi-step agent task that reads a codebase, runs tools, and writes hundreds of lines, while another “request” is a two-word question. Tokens measure the information a model processes and generates, so they track actual work much more closely than message counts do — at the cost of being harder for a user to predict in advance.

That trade-off is the real story of the December change. OpenAI is moving its subscriptions towards a meter that is fairer in the sense that it charges for work done, and less predictable in the sense that a user cannot easily tell, before starting a task, how much of their monthly pool it will consume. The Settings dashboard is the company’s answer to that unpredictability: one place to see the combined balance and the reset time. Whether that dashboard, plus the Instant-chat exemption, is enough to keep the change feeling like a subscription rather than a meter, will be decided by the allowance sizes OpenAI has not yet published — and by how subscribers actually experience the first shared-budget month.

Watch the Help Center article and the Settings screen in December. The exact day, and the numbers, will appear there first — and this time, the support page is the announcement.

For an explainer on how AI agents work, see AI Agents Explained: What They Are, How They Work & Uses, and for the tools shaping this market, see Best AI Tools 2026: What’s Worth Paying For and What to Skip.

Sources

Similar Posts