If your developers hold ChatGPT Plus, Pro or Business seats, their chat and their Codex work will draw on one token allowance from December, and OpenAI has not said how large that allowance is. You have about a month to measure current usage and decide what you will pay when a seat runs out.
OpenAI's Help Center article, updated October 31, 2026, says: "Starting in December, Plus, Pro, and Business plans' usage allowance will be shared across ChatGPT and Codex". It also says "Instant chat will not draw from your usage." The Help Center page returned an error when we tried to open it on November 1, 2026, so the wording here comes from write-ups that reproduce it. Read it yourself before you commit a budget.
What Changes in December for Plus, Pro and Business Seats?
Ordinary chat on advanced models joins a pool that already exists for agent work. OpenAI's current pricing documentation says "ChatGPT Work and Codex share usage". Today, regular chat has its own limits, so a long Codex session does not reduce what a developer has left for a question in ChatGPT. From December it will.
A token is the unit of information a model processes and generates, and it becomes the measure for both products. OpenAI says: "More advanced models and complex tasks generally use more of your allowance." Remaining usage and reset times will appear together under Settings > Usage. At the limit, a user can wait for a reset, upgrade or buy credits for eligible features.
The strongest case for the change is OpenAI's own. It says it expects most users won't need to wait for resets more often than they do now, and a developer who splits time between chat and Codex could come out ahead if capacity left over in one product becomes available to the other. ByteIota's analysis puts the loss on someone running four to six hours of Codex a day. An expectation about most users tells you nothing about your heaviest engineers, and those are the seats that stop work when they run dry.
Reaction so far is hostile: Reddit threads drew hundreds of upvotes, and some subscribers threatened to cancel Plus.
What Has OpenAI Not Published?
OpenAI has given no number you can budget against. TokenPost reports no launch day and no per-plan allowance amounts, and how chat and Codex tokens convert into the shared pool is also undisclosed. ByteIota notes there is no comparison with today's combined limits, so nobody outside OpenAI can say whether the pool is larger, smaller or the same.
Four more gaps affect how you plan.
Plus and Standard Business carry five-hour estimates, Pro has no five-hour limit, and weekly limits may apply. Whether the five-hour window, the weekly cap or both survive is open. StackVersus points out that OpenAI's plan pages do not yet match the shared model.
Credit-priced Business seats already pay token rates for Work and Codex, and how December changes them is not stated. OpenAI stopped offering pay-as-you-go Codex-only seats to new Business signups on June 24, 2026, while existing ones keep working, so check which kind you hold.
Enterprise and Edu are unclear. The quoted text names only Plus, Pro and Business. BigGo Finance's plan table marks Enterprise and Edu as outside the pool without giving a reason, while XenoSpectrum says the guidance does not address them. Get the answer in writing from your account team.
The current paid instant reset covers individual Plus and Pro accounts, not Business, and whether it lasts past December is unknown. Support does not reset limits, so a Business seat at its limit has fewer exits than a personal one.
Why Your October Numbers May Flatter Your Baseline
Usage data from the last four weeks probably understates how tight December will feel. BigGo Finance reports that Codex lead Thibault Sottiaux ran a 28-day campaign from October 5 with more than ten resets, and that Pro tiers and Business Premium had their limits doubled on October 29. CryptoBriefing describes the same move differently: Pro 100 and Pro 500 doubled, and a planned Pro 200 cut was cancelled. Neither report traces to an OpenAI page we could open, so treat the details as reported.
BigGo also quotes Sottiaux describing a move "from The Reset Company to The Token Company." We found that line in one source only. If it holds, a developer who never hit a wall in October may have been rescued by a reset, and standard Business seats may not have had the doubling at all. Record token consumption per person and per week, and mark the reset dates against it.
How Much Does One Coding Task Cost on the Current Rate Card?
The same token count can cost 100 times more or less depending on the model a developer picks. OpenAI's current credit rate card prices GPT-6 Astra at 100 times GPT-6 Luna, on both input and output. OpenAI has not said the shared pool will use these rates, so treat them as the best available proxy.
A credit is the unit OpenAI bills overflow usage in. Credits reportedly sell at 1,000 for $40, or four cents each, a figure CloudZero also reaches. OpenAI's docs do not state a dollar rate, and the price can depend on plan. Four cents does reproduce the $1 input and $5 output list price for GPT-6.1 Terra from Terra's 25 and 125 credits.
Here is our arithmetic for one illustrative task of 200,000 input tokens and 20,000 output tokens, at standard speed with no caching. Rates were read on November 1, 2026.
| Model | Credits per 1M input / output | Credits for the task | Cost at $0.04 |
|---|---|---|---|
| GPT-6 Astra | 250 / 1,250 | 75 | $3.00 |
| GPT-6.1 Sol | 50 / 250 | 15 | $0.60 |
| GPT-6.1 Terra | 25 / 125 | 7.5 | $0.30 |
| GPT-6 Luna | 2.5 / 12.5 | 0.75 | $0.03 |
Twenty Sol tasks a day over 21 working days comes to 6,300 credits, or about $252 a month per developer if all of it ran on purchased credits. Included allowance and cached input would reduce that. Treat it as a heavy case: OpenAI's pricing page says a typical GPT-5.6 Sol task may use 2-15 credits, so our 15-credit task sits at the top of the range it gives for that older model, and it gives no range for GPT-6.1 Sol. The monthly figure also sits above the $100 to $200 per active developer per month that CloudZero attributes to OpenAI, and your own task sizes will move it.
Fast mode costs 2.5 times Standard against included usage and 2 times against purchased credits. A team that leaves it on as a habit will burn the shared pool faster, and from December that also eats chat.
Which Way Out Should a Heavy Codex User Take?
Three exits exist at the limit, and the right one depends on the seat. Moving a person up a tier fixes a known heavy user when the published allowance justifies it, but Business lists at $20 per user per month billed annually or $25 billed monthly, with Pro starting at $100 a month, and you cannot compare them until OpenAI publishes sizes.
Workspace credits suit bursty demand. Business workspaces on flexible pricing can buy them. WorkOS reports that credits last 12 months and are non-refundable, and argues the rates match API list prices, so buy in small batches while the conversion rules are unknown.
An API key separates the heaviest terminal work from the pool altogether. Codex runs on an API key through the CLI, SDK or IDE at standard API rates, without cloud features. WorkOS says that costs you GitHub code review and the Slack integration, and its own benchmark found the credit lane about 1.5 times slower than the API. That benchmark is one vendor's measurement, so run your own before relying on it.
For routine questions, steer people to Instant, the one mode OpenAI has excluded. XenoSpectrum cautions that Instant is not unlimited. Any other chat should be assumed to count, since how non-Instant modes are counted has not been spelled out.
What to Do Before the Switch
This Week:
- Export current usage for your 10 heaviest Codex users from the Usage panel or
/statusin the CLI, and mark October's reset dates beside it. - Email your OpenAI account contact with five questions: the switch date, the allowance per seat type, the chat-to-Codex conversion, which windows survive, and whether your Enterprise or credit-priced seats are in scope. Ask for the answers in writing.
- Name one person who owns the decision to buy workspace credits, and give that person a monthly ceiling.
This Month:
- Set a default model policy with your engineering leads: Terra or Luna for light work, Sol for demanding tasks, Astra by exception, Fast off unless someone asks.
- Sort every seat into Instant-heavy, mixed or Codex-heavy, using the export. Only the third group needs a tier or billing decision.
- Move one Codex-heavy developer's terminal work to an API key for two weeks and record what the cloud features were worth to them.
Before the December Switch:
- Check the Help Center article and the pricing page each week, and rerun your cost model the day allowance sizes appear.
- Tell developers in advance that a long Codex run can now leave them short for chat, and where to see remaining usage.
The Bottom Line
Seat plans for AI tools keep moving toward metering. We have covered the same turn at GitHub Copilot and Microsoft 365 Copilot, and in both the price on the seat stopped telling a buyer what the seat would cost. ChatGPT Business now heads the same way, with less published detail than a finance team would want this close to the date. Our cap-first guidance on consumption pricing applies here too.
OpenAI may publish generous allowances, and many mixed users may never notice the change. You still need the baseline, because it is the only way to tell afterwards which of those happened to you. Send the five questions to your account contact this week.