ChatGPT Codex Shared Token Limit: December 2026 Guide

For months, ChatGPT subscribers lived with two separate meters. Chatting in ChatGPT was one thing; coding in Codex was another. A heavy week of agentic coding never ate into your chats, and a long evening of brainstorming never slowed your coding. From December, that separation reportedly ends — and the meter itself changes from messages to tokens.

According to an updated OpenAI support page, reported by The Decoder on 1 November 2026, "starting in December, Plus, Pro, and Business plans will share a single token budget across ChatGPT and Codex" as OpenAI folds Work's agentic features into the main ChatGPT interface. The same wording has been quoted by other outlets reading the same help-center article. Two honest caveats belong right up front: OpenAI has not published a standalone announcement of the change, and no exact December date has been given. A tracker that checked OpenAI's help and pricing pages on 30 October also noted staff saying the Chat/Work merger was "not rolling out just yet." So read everything below as the reported shape of the change, not as OpenAI's final, confirmed terms.

That distinction matters, because this is a subscription decision guide, not a rumour round-up. What follows sticks to what the sources actually say, flags what is still unknown, and gives you one concrete check to run before December.

What actually changes

Today, usage in ChatGPT and usage in Codex are metered separately for subscribers. From December, the reported change pools them: one token-based allowance, drawn on by both ChatGPT and Codex. Use a lot of Codex on Monday and Tuesday, and there is less left for ChatGPT on Wednesday — and the other way round.

Three details from the reported support-page wording define the new system:

• The meter is tokens, not messages. Tokens are the units of information a model processes and generates. A short factual question costs a handful; a long reasoning run, a big file analysis, or an agentic coding session costs far more.
• The model you pick sets the price. OpenAI's own wording, as quoted in the reporting, says more advanced models and complex tasks "generally use more of your allowance." Choosing a flagship reasoning model for a trivial question is, under a token budget, the most expensive habit you can have.
• Instant chat is exempt. "Instant chat will not draw from your usage," the support page reportedly says. Instant is the fast default model for quick replies — so the lightest kind of use stays off the meter entirely.

When the pool runs out, the reported options are to wait for the reset, upgrade to a higher plan, or buy extra credits. Your remaining balance and reset times are shown in Settings → Usage, and that screen is the single most useful thing in this whole story.

Why the timing stings for Plus users in particular

The December change does not land on a level playing field. On 30 October, OpenAI doubled usage limits for its Pro 100, 200 and 500 plans and for Business Premium, and scrapped a planned cut to the Pro 200 allowance after user pushback. Plus limits, by contrast, were left unchanged.

That asymmetry is the real story for most readers. Plus costs $20 a month; Pro starts at $100 a month, with $200 and $500 tiers above it. Pro plans currently have no five-hour limit in Work and Codex, while Plus does. So the subscribers with the smallest allowance are also the ones whose allowance was not enlarged ahead of the merger — and they are the group most likely to feel two previously separate meters becoming one.

None of this means Plus becomes a bad deal. It means Plus subscribers who also use Codex are the group that should actually check their usage, while everyone else can probably ignore the whole story.

How much does a task really cost? The honest answer: it depends on the model

OpenAI has not published a per-tier token allotment for December, and no source in this story does either. Any article quoting a December token number for Plus or Pro is inventing it. What OpenAI does publish is estimates for current Work and Codex usage on Plus — and those estimates show why model choice matters more than message count.

For a five-hour window on Plus, OpenAI estimates roughly 15 to 160 local messages with GPT-6.1 Sol, depending on the task. Other coverage of the same help-center estimates puts the spread even wider across models: a handful of messages with the most advanced model versus hundreds or thousands with a lighter one in the same window. These are today's estimates for Work and Codex, not December figures — but they illustrate the mechanism exactly. Under a token budget, "how many messages did I send" stops being the right question. The right question is "which model did I send them to, and how hard did I make it work."

The check to run before December

Do not guess, and do not upgrade on a headline. Open ChatGPT, go to Settings, then Usage, and look at two things: your remaining balance and your reset times. Then ask yourself three questions:

1. Do I hit my limit in a normal week? If the answer is no, December changes nothing for you. Close the tab.
2. Where does my usage actually go? If most of it is Codex, the merger is what matters to you. If most of it is long reasoning chats, the token meter itself is what matters — try the lighter model for routine work and save the flagship for work that earns it.
3. What does the limit cost me when I hit it? A few hours of waiting is free. Extra credits cost money but keep your plan. Pro costs $80 a month more than Plus at its lowest tier. The right answer depends on how often you hit the wall, and only your Usage screen can tell you that.

A few practical habits also get more valuable under a token budget, and none of them requires spending anything: use Instant for quick questions; keep heavy reasoning for genuinely hard problems; break huge Codex tasks into smaller runs rather than one marathon session; and clear the decks before a big coding day if you also need chat allowance later in the week.

What is still unknown — and why we are saying so

A subscription guide that hides its gaps is worse than useless, because readers make money decisions from it. Here is what remains genuinely uncertain at the time of writing:

• The exact December date. None has been published.
• Per-tier token allotments. None has been published. Do not trust any number you see quoted for December.
• The price of extra credits. Mentioned as an option, not priced in the reported wording.
• Whether the merger itself slips. Staff comments reported on 30 October suggested the rollout was not imminent, even as the support page carried December wording. Both facts can be true — documentation can be ready before the rollout is — but only OpenAI can resolve the tension.

The sensible posture is to treat December as the planning assumption and your Usage screen as the ground truth. If OpenAI announces firm terms, this guide's advice narrows to a number; until then, "check your usage" is the whole strategy.

The bottom line

The December change, as reported, is not the end of flat-rate ChatGPT — Instant chats stay outside the meter, and light users will notice nothing. It is the end of pretending that chats and coding draw on different wallets. For heavy Plus users who do both, that is a real change to plan around, and the plan costs nothing: check Settings → Usage, see which meter you actually drain, and let that — not a headline, and not an upgrade button — decide what you do next.

If you want the wider context on which assistant fits your work in the first place, see ChatGPT vs Gemini vs Claude in 2026. For how Codex usage already works on the model side, see OpenAI GPT-6.1 Terra explained and the best AI coding tools for small teams in 2026. And for a reminder that a different ChatGPT feature — share links — carries its own separate risk, read today's report on the fixed share-link flaw.

Recommended Reading

If the economics behind these pricing shifts interest you, Kai-Fu Lee's AI Superpowers is a useful companion read — it compares the US and Chinese AI ecosystems and what the AI economy means for jobs. Read more on AI Superpowers at BookNest.

As an Amazon Associate, AI Trends Daily earns from qualifying purchases.

The featured image is an original AI-generated illustration, not a photograph. Sources: The Decoder, "OpenAI's upcoming chat-and-work merger in ChatGPT signals the end of all-you-can-prompt pricing" (1 November 2026); FourWeekMBA, "ChatGPT and Codex Will Share One Usage Allowance in December". Plan prices are OpenAI's list prices as reported in October 2026 coverage of its pricing pages; check the live pricing page before subscribing, as tiers and allowances change often.