ChatGPT and Codex Will Share One Allowance in December: A Single Token Wallet for Everything
Starting in December, your ChatGPT allowance and your Codex allowance are the same one, counted in tokens: a long coding session can leave you no room to ask anything. And OpenAI still hasn't said how big each plan's pool is.
What OpenAI Has Confirmed
The company updated its help page on usage limits. What's confirmed fits in a few sentences, and it's worth separating from everything else, because that's almost all the hard information there is today.
One Allowance for Plus, Pro and Business
From December, on the Plus, Pro and Business plans the usage allowance is shared between ChatGPT and Codex. It isn't a price change: it's a change of meter. The three plans cost the same as before.
Measured in Tokens, Not Messages or Hours
Usage will be measured in tokens, the units a model processes and generates. How much you spend depends on the model you pick and the task you ask for: a short question on a light model looks nothing like a long reasoning session. That's why the same plan can last very differently depending on how you use it.
Read also
Instant Chat Doesn't Draw From That Pool
Instant mode stays outside the shared wallet, and OpenAI describes it as unlimited on paid plans. The fine print is in the rest: higher-reasoning conversations do appear to consume from the common pool. The quick question stays free; thinking slowly does not.
The Reason: Work's Agent Capabilities Move Into ChatGPT
The company's stated reason is that it is bringing Work's agent capabilities into the main ChatGPT experience. In plain terms: if chat, the agent and the code editor are going to be one product, they have to spend from one meter. Keeping three separate allowances for something presented as a single product doesn't hold up for long.
What It Hasn't Confirmed
Here's the practical problem, and it's a serious one for anyone who has to budget.
How Many Tokens Each Plan Gets
There are no figures. No tokens per plan, no equivalence with today's limits. Nobody knows whether the pool will be the same, bigger or smaller than what you have now. Any number you see on social media is a third-party estimate, not official data.
Which Day in December It Starts
The notice says "December" and nothing more. There's no date. If you need to plan a project deadline around it, the information simply doesn't exist yet.
Whether the 5-Hour Window and the Weekly Cap Survive
Today, agent usage is split across a five-hour window and a weekly cap. The notice doesn't say whether those two clocks survive the change. It's an open question that changes the experience considerably: a daily ceiling and a monthly one are not the same thing.
Why OpenAI Is Doing This
One Product, One Meter
Merging the Work tab into chat points the same way: one surface, one history, one allowance. As product design it's coherent. As a billing arrangement, it means two very different activities compete for the same resource.
The Real Cost Comes From the Agent, Not the Single Question
A question in chat costs little. An agent task that reads files, runs commands and tests code costs far more: that's many model calls inside a single job. As those features become the center of the product, spending concentrates there. A single meter makes that cost visible instead of hiding it in a separate allowance.
The Precedent: Work and Codex Already Shared a Pool
According to OpenAI's own Codex documentation, work started from Work or from Codex on desktop already uses the same pool, with its five-hour window and weekly cap, while regular chat doesn't touch it. What changes in December is that ChatGPT joins that same pool for the three paid plans. The direction was already set.
Why Developers Are Worried
A Long Codex Session Can Eat the Month
With a single shared balance, an intense day of agent-assisted coding can leave you without room for the rest of the week. And the reverse: heavy reasoning chat comes out of what you'd set aside for code. That's the inevitable consequence of one wallet, and OpenAI already warns that some users will need to wait for resets more often than usual.
Chat Stops Being Effectively Free
Instant mode is exempt, so the basics stay covered. But serious ChatGPT work — the kind that uses long reasoning and tools — stops being a separate resource. For anyone who treats chat as a working notebook, that's the change they'll feel.
What the Community Wants: Publish the Numbers, Keep the Option to Split
On OpenAI's official forum there are threads asking the company to publish per-plan balances before December and to keep the option of separate limits. It isn't a complaint about price, it's about the missing figures: without numbers you can't decide whether to change plan, change tool or change method.
How to Prepare Before December
Measure Cost per Task Before They Measure It for You
The only real defense against a new meter is your own measurement. Write down what a typical task costs you: one refactor, one test, one code review. If you don't know what you spend today, the change will surprise you regardless of how big the pool turns out to be.
Cheap Model for Routine Work, Prompt Caching for Repetition
Much of the consumption goes to mechanical tasks that don't need the most expensive model. Dropping to a cheaper model for repetitive work and caching the prompt that repeats are the two levers with the fastest effect. We already covered how to track spend per task and cache prompts with code, and the method applies here unchanged.
Automation Through the API, Not Through the Subscription
If you have processes that run on their own, the subscription was never the right place: there you pay for actual usage with no plan ceiling. Watch out for the confusion, because there are two separate wallets: the API credits that Pro and Business started including in late October are API balance that expires each cycle, not your subscription allowance.
A Budget per Project and an Alert Before the 25th
Set a usage ceiling for each project and check progress mid-month, not on the last day. A meter change mid-cycle is far easier to manage with two weeks of margin than with two days.
What Not to Do
Don't Assume Today's Allowance Carries Over
It's the most expensive mistake and the easiest one to make. What lasts you a month today may last considerably less, and there is no published data saying otherwise.
Don't Move Everything to the API on Day One Without Numbers
The API charges for actual usage. If your usage is high and steady it usually works out better; if it's occasional, it can be worse than the subscription. Without measuring first, migrating is a bet.
Don't Upgrade Out of Fear of a Number Nobody Has Published
With no published figures, paying for a higher plan is paying for an unknown. Wait for the official numbers: if the balance falls short, you can always upgrade afterwards.
Conclusion
No tool disappears: the meter changes. Measure your usage before December and you keep the bill under control; skip it and you find out mid-month, with the allowance already gone. We pay for these subscriptions and measure them on this blog, so we'll report what each plan actually includes once OpenAI publishes the numbers. If the real cost of AI in daily work interests you, keep reading.
