In December, ChatGPT and Codex start drawing from a single allowance on Plus, Pro, and Business plans. The unit changes from messages to tokens, so the same question can cost very different amounts depending on how you ask it.
What OpenAI Changed for Plus, Pro, and Business
OpenAI's help center now says usage limits for Plus, Pro, and Business will be shared between ChatGPT and Codex starting in December. The company ties the change to folding Work's agentic features into the main ChatGPT experience.
Agentic, in plain terms, means software that takes a series of steps on its own instead of answering one prompt. Codex writes and runs code across many steps. Work does similar things with documents and tasks. Once both sit inside ChatGPT, one meter makes more sense to OpenAI than two.
The change covers three plans. OpenAI hasn't described a change for Free.
Tokens Are a Different Unit Than Messages
A message is countable. You send one, it counts as one, and the cap is easy to understand even when it's annoying.
A token is a piece of text, roughly a word fragment. Models read and write in tokens, and the count depends on what goes in and what comes out. OpenAI says how much you use depends on the model you pick and the type of task, with stronger models and harder tasks drawing more.
That's the part worth sitting with. Two people on the same plan can send the same number of messages and land in very different places. A Codex run that reads a dozen files and writes a patch isn't the same size as a one-line chat reply.
The old system hid that difference. Message caps made a long reasoning session and a quick lookup cost the same. Token counting puts the difference on the bill.
Instant Chat Still Doesn't Touch the Pool
One exception is explicit in the help center: Instant chat will not draw from your usage. Stay in the fast default mode and the merged allowance doesn't move.
Everything else is in scope. That includes the heavier chat models and Codex sessions. A morning of refactoring with Codex can consume what a week of casual chatting wouldn't.
If your ChatGPT habit is mostly Instant, the change may never show up in your day. If you switch models for hard questions, it will.
Where to Find the Combined Number
OpenAI points users to Settings, then Usage. That screen shows the merged remaining allowance and the reset time.
Two things are worth checking there once December arrives. Whether the reset clock is shared or each product keeps its own cycle. And whether the number you see is a percentage or an absolute token count.
OpenAI's description talks about one allowance and one reset. That suggests a single clock. The help center doesn't spell out the edge cases, so the screen is the authority once it ships.
Three Questions OpenAI Left Open
The help center answers the "what" and skips most of the "how much."
Confirmed
- Shared allowance across ChatGPT and Codex on Plus, Pro, and Business
- Token as the counting unit
- Instant chat exempt
- Settings, then Usage, shows the merged balance and reset time
Still unanswered
- The exact start date in December
- Token amounts for each plan
- How existing message caps convert
None of those gaps is small. Without token amounts per plan, nobody can say whether their current usage fits. Without a conversion rule, nobody can compare last month's limits with next month's.
Watch for the numbers rather than the announcement. A shared pool with a generous total feels like an upgrade. The same pool with a tight total feels like a cut, and the help center doesn't let you tell which one is coming.
What This Means If Codex Is Your Main Tool
Heavy Codex users have the most to lose from a shared pool, at least at first. A long agent session can eat a large share of a monthly allowance, and it does so in bursts rather than evenly across the month.
The reverse is also true. Someone who mostly chats and occasionally runs Codex may never notice.
Business admins have a separate problem. Team quotas were probably set with chat in mind. If a few developers start running Codex all day, the pool drains faster than the old model predicted, and the help center describes no per-person lever for the admin.
That's the gap to raise with OpenAI if you manage a team. A shared pool without per-seat controls turns one person's heavy week into everyone's problem.
What OpenAI Says About Disruption
OpenAI says it's preparing for the switch to keep workflow interruptions small. It also says it expects most users won't need to wait for a reset more often than they do now.
That's the company's own forecast. No outside measurement backs it yet, and OpenAI hasn't published the modeling behind it.
Treat it as a statement of intent about the transition. It isn't a guarantee about your account, your model mix, or your team.
How to Get Ready Before December
Three practical moves, none of them dramatic.
Check Settings, then Usage, once a week between now and December. You want a sense of what a normal week looks like before the units change, so the new number means something when it appears.
Then list the habits that burn the most: long Codex runs, large file reads, regenerating the same output five times. Those are the ones that will show up in a merged pool first.
If you're on Business, ask your admin whether quotas get re-cut. If you're on Plus or Pro and you run Codex daily, the three options OpenAI lists are worth knowing now: wait for the reset, upgrade, or buy credits for eligible features.
That's the picture as published. The date, the token amounts, and the conversion rule will decide how it feels in practice.






