ChatGPT and Codex Share One Limit in December: What Changes

If you pay for ChatGPT and you also use Codex to write code, your two separate usage budgets are about to become one. OpenAI has confirmed that, starting in December, Plus, Pro and Business plans will share a single usage allowance across ChatGPT and Codex, measured in tokens rather than messages. A long afternoon of coding and a long evening of chatting will draw from the same pool — and only one kind of chat is exempt.

Quick answer: From December 2026, ChatGPT and Codex share one token-based usage allowance on Plus, Pro and Business plans. How fast it drains depends on the model you pick and the task you run; heavier models and harder tasks use more. Instant chat will not draw from your usage. You can check your combined remaining usage and reset times in Settings → Usage, and if you hit the limit you can wait for a reset, upgrade your plan, or buy credits for eligible features. OpenAI has not published an exact start date or the size of any plan's allowance.

Contents

• What OpenAI actually announced
• How the token-based limit works
• Before vs. after: what changes in practice
• Who is affected — and who is not
• The second change: card and device limits
• How to prepare before December
• Pros and cons: an honest read
• Who should worry, and who can relax
• Frequently asked questions
• Our verdict

What OpenAI actually announced

The change arrived quietly. There was no launch blog post and no press release. Instead, OpenAI updated its Help Center article, Understanding ChatGPT and Codex usage limits, which surfaced on October 31, 2026 and was widely reported on November 1. The article is short, and every sentence in it matters, because it is the only official word so far.

Here is the core of it, in OpenAI's own words: "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." The reason given is product consolidation. ChatGPT Work — the longer, multi-step, deliverable-focused mode — already shares an allowance with Codex. December pulls ordinary ChatGPT conversations into that same shared pool.

Two other sentences deserve the same attention. "Your usage for ChatGPT and Codex will be measured in tokens, the units of information a model processes and generates." And: "Instant chat will not draw from your usage." That last line is the one exemption OpenAI names, and it is the single most useful detail for anyone trying to stretch an allowance.

OpenAI also sets an expectation, carefully worded: "We expect most users won't need to wait for their usage to reset more often than they do now." Treat that as a forecast, not a promise. It ships with no token figures, no per-plan allowance sizes, and no reset schedule, so there is currently no way to compare your December allowance with what you have today.

How the token-based limit works

Until now, most ChatGPT users thought in messages: so many messages per window, per model, per week. Tokens work differently. A token is a chunk of text — roughly part of a word — that the model reads in and writes out. A short question costs a handful of tokens. A long document you paste in, a long reasoning chain, or a coding task that reads a whole repository costs far more, even though each one might count as a single "message" under the old way of thinking.

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 is the whole pricing logic in one line: the model picker becomes a budget decision, not just a quality decision.

You can already see the shape of this in how Work and Codex are metered today. OpenAI's usage guidance for Work and Codex says your allowance does not correspond to a fixed number of messages or tasks, and it publishes estimated ranges for a five-hour window on Plus — estimates, not guarantees, and they vary with the task, model and settings. The gap between the cheapest and the most capable model is enormous:

Model (Work and Codex, Plus, 5-hour window)	Estimated local messages
GPT-6 Astra	5–45
GPT-6.1 Sol	15–160
GPT-6 Sol	15–150
GPT-6 Luna	350–3,000

Those numbers come from OpenAI's own usage guidance for GPT-6 Astra in Work and Codex, and they describe the current Work and Codex pool — not the December ChatGPT pool, whose sizes OpenAI has not published. Still, they make the point clearly: picking the heaviest model for every question is the fastest way to burn an allowance, and picking a lighter model for routine work is the easiest way to save one. That trade-off is worth understanding now, because from December it will apply to everyday chat too. If you want to see how the newest ChatGPT models actually behave in regular conversations, our walkthrough of GPT-6 with Intelligent UI in ChatGPT is a useful companion read.

What counts, and what does not

• Counts: ChatGPT conversations (other than Instant chat) and Codex work, on Plus, Pro and Business plans, from December. Work, Codex, ChatGPT for Excel and Workspace Agents already share one allowance and credit pool where your plan includes them.
• Does not count: Instant chat. OpenAI states this plainly: "Instant chat will not draw from your usage." That makes it the natural home for quick, everyday questions that do not need deep reasoning.
• Unknown: Whether the per-model message caps in chat — such as Pro's weekly GPT-6 Pro message allowance — stay alongside the new pool or disappear into it. The December article does not say. The same goes for the five-hour and weekly windows that govern Work and Codex today: OpenAI has not said whether those windows change.

Where to check your usage

From December, you will see "your combined remaining usage and reset times in Settings → Usage." That is already the place to look today for Work and Codex usage, so the habit transfers directly. If you use Codex from the command line, the /status command in an active session shows which allowance is exhausted, your credit balance and any reset time. Start checking that screen weekly now, while the two pools are still separate, so you have a baseline for what your normal ChatGPT use and your normal Codex use each look like on their own.

Before vs. after: what changes in practice

	Today (before December)	From December
ChatGPT chats	Separate message-based limits	Draw from the shared token allowance
Codex / Work	Shared allowance between Work and Codex	Same pool, now joined by chat
Metering unit	Mostly messages and windows	Tokens, varying by model and task
Instant chat	Lightweight everyday chat	Still exempt — does not draw from usage
Usage screen	Settings → Usage (Work/Codex)	Settings → Usage, combined balance and resets
At the limit	Wait, upgrade, or buy credits	Same three options, for eligible features

The practical consequence is simple to state and easy to underestimate. Before December, a heavy Codex session could not eat into your chat allowance, and a heavy chat day could not eat into your coding allowance. From December, they compete. A long agent run that reads a large codebase, reasons for a while and writes a lot of code will leave less room for chat that day — and a day of long, heavy-model conversations will leave less room for Codex that evening. The order matters less than the total: it is one budget now.

That is also why the December article's silence on allowance sizes matters so much. If the shared pool is sized generously, most people will notice nothing, exactly as OpenAI predicts. If it is sized close to today's Codex pool alone, people who use both products heavily will notice it quickly. Nobody outside OpenAI can tell yet, which is why the honest advice for November is to measure your own use rather than guess.

Who is affected — and who is not

The December article names exactly three plan families: Plus, Pro and Business. If you are on one of those, the shared allowance applies to you. If you are on Free, Enterprise or Edu, the article does not mention your plan at all. That is not the same as saying nothing changes for you — it means OpenAI has said nothing about you yet, and you should watch for a follow-up rather than assume either way.

Within the affected plans, the impact is uneven by design. Someone who chats a little and never opens Codex will feel almost nothing: their chat use simply moves to a new meter, and OpenAI expects most such users will not hit resets more often. Someone who lives in Codex all day and chats in between — a working developer, a freelancer shipping client work, a small team running research and code in the same account — is the person this change is really about. Our guides to the best AI tools for freelancers and the best AI tools for small business both assume ChatGPT sits at the centre of a working day; for those readers, the December pool is a real planning item, not a footnote.

Model choice sharpens the divide further. GPT-6.1 Terra, the speed-focused model, and the lighter Luna-class models are built to be cheap to run; Astra-class reasoning is built to be thorough, and thorough costs tokens. The sensible default from December is the one that already makes sense today: light models for routine work, heavy models when the task genuinely needs them, and Instant chat for the quick questions in between.

The second change: card and device limits

The shared allowance was not the only restriction OpenAI documented in the same few days. A second Help Center article, Understanding card and device limits in ChatGPT, adds caps on how many paid accounts can share one payment card or one device — and this one is already live, with sharper edges.

There are two limits, and they bite at different moments. The card limit works at checkout: if too many subscriptions are associated with the same payment card or device, ChatGPT may block a new purchase with the message, "Too many subscriptions associated with this card or device. We can't complete this purchase." OpenAI is explicit that this is different from a bank decline — the payment did not fail, the purchase was blocked. The device limit works at sign-in and after it: if you see "Device limit reached," you cannot use that paid account on that device. You may see the warning when you sign in, or you may be signed out after a later check, "even if the purchase went through and the account worked at first."

The plans covered are ChatGPT Go, Plus, Pro 100, Pro 200 and Pro 500. Free accounts and Business, Enterprise and Edu workspace plans do not count toward the device limit. OpenAI has published no numbers for either cap — no threshold, no count, no breakdown of how the limit is calculated. That missing number is the most frustrating part of the policy, because there is no way to know in advance whether you are one account away from the line.

Three details catch people out. First, signing out does not fix it: OpenAI says signing out of an account "does not guarantee that another paid account becomes eligible." Second, buying credits or resetting usage will not clear a card or device restriction — those fix usage limits, which are a different thing. Third, the limits can catch innocent setups: if several people use the same payment card, the card can reach its checkout limit even when each person uses a different device, and a shared household or public device can reach its limit when people use their own accounts. OpenAI's own help page concedes both cases and tells affected users to mention the shared card or device when contacting Support.

OpenAI frames both limits as "part of its privacy and security protections." The Decoder, which first reported the change on November 1, reads it differently: as a push away from stacking several cheaper subscriptions to squeeze out more usage, and toward buying extra credits or usage resets, which it notes are generally more expensive than another subscription. That is The Decoder's interpretation, not OpenAI's stated reason — but either way, the practical effect is the same. The era of rotating multiple paid accounts on one device to multiply a flat-rate allowance is closing, and the sanctioned alternative OpenAI points to is credits or resets on an account you already own.

One warning from the help page deserves repeating verbatim in spirit: subscriptions bought before these restrictions take effect can continue under an existing-subscription exception, but cancelling one "does not guarantee you can replace it." OpenAI advises contacting Support before making changes if you are unsure. If you do hit a device warning, the options OpenAI lists are to sign in with another permitted account you own, or to use the cancellation-and-refund option shown in the warning — for ChatGPT or Google Play purchases, that ends paid access immediately and starts a refund, and it may affect multiple subscriptions on the account. Do not cancel in a hurry and assume you can simply buy the same plan back.

Why is OpenAI tightening usage now?

OpenAI has not linked the two changes to any single cause, and its stated reason for the shared allowance is product consolidation: Work's agentic capabilities are moving into the main ChatGPT experience. The wider context, reported separately, is a cooler funding climate. The Financial Times reported on October 30 that OpenAI had cut its latest fundraising target from about $30 billion to about $20 billion, at a $1.25 trillion valuation down from $1.4 trillion, with the Wall Street Journal reporting about $11 billion committed so far. That reporting is about investor appetite, not about your subscription — but it is the backdrop against which every usage policy now gets read. The honest summary is that OpenAI is moving, at the same time, toward metering usage more precisely and toward closing the workarounds that let a single flat fee stretch further. Whether that is discipline, consolidation, or both, the user-facing result is the same: usage is becoming something to manage, not something to assume.

How to prepare before December

None of this needs code, and none of it needs a new subscription. It needs a week of paying attention, before the meter changes.

1. Check Settings → Usage now. Look at your remaining allowance and reset times for Work and Codex, and note how often you actually hit a reset today. That is your baseline.
2. Separate your two habits for a week. Notice which days are chat-heavy and which are Codex-heavy. If they rarely overlap, December will probably feel like nothing. If they collide most days, you are the person who needs a plan.
3. Move quick questions to Instant chat. It is exempt from the shared pool, so it is the obvious home for the short, everyday questions that currently sit on a heavier model out of habit.
4. Match the model to the task. Use lighter models for drafting, summarising and routine code, and save the heaviest reasoning for the tasks that genuinely need it. The model picker is now a budget control.
5. Write better prompts before you run them. A precise prompt that gets the answer in one run costs far less than three vague runs that circle the same problem. Our free AI Prompt Generator can help shape a sharper first prompt for coding and research tasks alike.
6. Batch Codex work. Give Codex a clear, complete brief in one task rather than a drip of small follow-ups. Long-running tasks that keep re-reading context are among the most expensive things you can run.
7. Know your fallback before you need it. If you hit the limit, your options are to wait for the reset, upgrade, or buy credits for eligible features. Decide in advance which one you would actually choose, so a mid-deadline limit does not become a mid-deadline panic.
8. Do not stack accounts to dodge it. The card and device limits are designed to stop exactly that, the numbers are unpublished, and cancelling a subscription to rearrange things may leave you unable to buy it back. One account, managed well, is the safe setup.

If you are comparing coding assistants rather than living inside Codex alone, the same budgeting logic applies everywhere. Our GitHub Copilot vs Cursor AI comparison walks through how the two main alternatives price and limit their agent work, and Anthropic's Claude Duet shows a rival taking a different approach to proactive agent help. None of them is free of limits either — the question, everywhere now, is which limit fits how you actually work.

Pros and cons: an honest read

What is genuinely better

• One screen, one truth. A single combined balance in Settings → Usage is easier to understand than two separate budgets with different clocks.
• Tokens are fairer than messages. Under message counting, a one-line question and a half-hour reasoning run looked identical. Token metering charges roughly for what the work actually costs.
• Instant chat stays outside the meter. Everyday quick questions keep working without touching the budget that coding and deep work draw on.
• Work and ChatGPT finally behave as one product. If agentic features are moving into the main experience, one allowance for one experience is at least coherent.

What is worse, or simply unknown

• Heavy users of both products lose headroom. Two budgets becoming one is a cut for anyone who used both fully, unless the new pool is sized to match the sum — and OpenAI has not said it is.
• No numbers, no date. No allowance sizes, no exact December start day, no word on whether chat message caps survive. Planning around a policy with no figures in it is guesswork.
• The card and device caps are opaque. Unpublished thresholds, late-appearing warnings, and no guarantee that cancelling lets you rebuy is a harsh combination, especially for shared devices and shared cards.
• Credits are the fallback, and credits cost. The sanctioned way to keep working past a limit is to pay more, which is exactly what flat-rate subscribers were trying to avoid.

Who should worry, and who can relax

Relax, mostly: if you use ChatGPT for everyday questions, occasional writing help and light research, and you rarely or never open Codex. Your use fits comfortably inside the pattern OpenAI says will not hit resets more often, and Instant chat covers much of it for free of the meter. Check Settings → Usage once, confirm you are nowhere near a reset, and carry on.

Pay attention: if you use Codex daily, run Work tasks for deliverables, or switch between chat and code all day on Plus. You are the user the change is built around. Measure your November use, move quick questions to Instant chat, and keep the model picker honest. You do not need to upgrade today — you need data, and November is when you can still collect it under the old rules.

Act now: if you currently run more than one paid ChatGPT account on a single device or card, or your household shares a card or a device across several paid accounts. The card and device limits are already documented, the thresholds are unpublished, and a warning can arrive after a purchase that worked at first. Read the help page, talk to Support before cancelling anything, and do not assume logging out frees a slot.

Frequently asked questions

When exactly does the ChatGPT and Codex shared limit start?

OpenAI says only "starting in December." No exact day in December 2026 has been published, so watch the Help Center article and the Settings → Usage screen for the change.

Does the shared limit apply to the Free plan?

OpenAI's December article names only Plus, Pro and Business plans. Free, Enterprise and Edu plans are not mentioned, so there is no confirmed change for them yet.

Will Instant chat really not use my allowance?

Yes, according to OpenAI: "Instant chat will not draw from your usage." Quick everyday questions on the Instant model stay outside the shared token pool.

How will I know how much allowance I have left?

OpenAI says you will be able to see your combined remaining usage and reset times in Settings → Usage. Codex command-line users can also run /status in an active session.

What happens when I reach the limit?

You can wait for your usage to reset, upgrade your plan, or buy credits to continue using eligible features. Credits pay for eligible usage after included limits are reached and do not transfer usage between accounts.

Can I just buy a second Plus subscription on another account?

Probably not safely. OpenAI now caps paid accounts per payment card and per device, and the exact thresholds are unpublished. A blocked purchase, or a "Device limit reached" warning, can follow — and buying credits or resetting usage will not clear that kind of restriction.

Our verdict

December's shared allowance is not a price rise, and it is not the end of flat-rate ChatGPT. It is something subtler: the moment ChatGPT's usage rules stop being something you can ignore and start being something you manage. For most people, that management will be light — Instant chat for quick questions, the right model for the job, a glance at Settings → Usage now and then. For developers and heavy users who live in both ChatGPT and Codex, it is a real constraint arriving without the numbers needed to plan for it, and OpenAI owes those users the allowance sizes before the switch flips.

The card and device limits are the harder pill. A policy that can lock a paid account out of a device, with unpublished thresholds and no guaranteed way back after cancelling, needs clearer numbers and clearer warnings than it has today. Until OpenAI publishes them, the safest advice is the simplest: one account, one card, one device where you can manage it — and Support before you cancel anything.

We will update this article when OpenAI publishes the December start date, the per-plan allowance sizes, or any change to the card and device thresholds. If you found this useful, the next question most readers ask is how the alternatives compare — start with our GitHub Copilot vs Cursor AI comparison, or see how GPT-6's Intelligent UI changes everyday ChatGPT use.

📚 Read next

• GPT-6 with Intelligent UI: What Changed in ChatGPT
• GitHub Copilot vs Cursor AI: Which Should You Pick?
• Claude Duet Launched: Anthropic's Proactive AI Agent Explained

Last updated: November 2, 2026. How we reported this: we read OpenAI's Help Center articles on ChatGPT and Codex usage limits and on card and device limits first-hand on November 2, 2026, and cross-checked the December change against independent coverage published November 1, 2026. Where OpenAI has published no figure — allowance sizes, the exact December date, card and device thresholds — this article says so rather than estimating. This article is not sponsored, and The AI Tools has no commercial relationship with OpenAI. Learn more on our About page or Contact page.