One Allowance, Three Plans: How OpenAI's Shared ChatGPT and Codex Budget Changes the Buying Math for Plus, Pro and Business
Metering layers rarely make headlines, yet they are where subscription economics actually live. OpenAI told users recently that, as the Chat and Work experiences inside ChatGPT are folded together, ChatGPT and Codex will draw from a single shared allowance within the coming weeks. The OpenAI Help Center dates the shift to December and names three affected subscriptions: Plus, Pro and Business. For readers whose working week is split between drafting, analysis and code, the practical question therefore moves fast from "what does this cost" to "which task wins when the pool runs short." Below is the verified mechanism, a plan-by-plan comparison, and the scenario verdicts we would act on today.
The facts, separated from the interpretation
- Scope: one shared usage allowance across ChatGPT and Codex, starting in December, on Plus, Pro and Business.
- Metering logic: consumption depends on the model selected and the task performed; higher-tier models and more complex work draw more.
- Exemption: chatting in Instant mode does not touch the shared allowance; OpenAI's notice states that current plans support unlimited use of Instant.
- Visibility: once the change lands, merged remaining allowance and reset timing sit in Settings, under Usage.
- When the pool empties: wait for the reset, move to a higher plan, or purchase additional credits to keep using eligible features.
- Interface work running in parallel: OpenAI plans to merge the Chat and Work tabs, first testing with a small group of users and widening the rollout on feedback; afterwards, people can discuss an idea and hand ChatGPT a task in the same view, for example building a site, producing a presentation or handling web work from supplied files and plugins, while still reviewing and steering the job in progress.
- What stays separate: Codex keeps a standalone experience inside the ChatGPT desktop app.
- Company expectation: OpenAI predicts that most users will not wait for allowance resets any more often than they do now.
Before and after: two clocks become one budget
The change is not a new restriction bolted onto existing ones. It removes the wall between two independently reset allowances and replaces coarse message counting with a resource-denominated one.
| Dimension | Prior arrangement | From December |
|---|---|---|
| Metering unit | Chat allowance historically expressed as how many messages or runs you could still send | Token-based accounting, weighted by model tier and task complexity |
| Reset clocks | Two: one for ChatGPT, another for Codex | One shared clock across both surfaces |
| Headroom transfer | Unused coding quota could not subsidise chat, and vice versa | Any task competes for the same headroom |
| Lightweight chat | Part of the ordinary conversation allowance | Excluded from the pool via Instant mode |
| Monitoring | Quota read separately per product | Single combined figure and reset time in Settings and Usage |
| Overage path | Product-specific: wait, or switch tool | Uniform menu of wait, upgrade, or top up with credits |
Context that is easy to miss: on the coding side, limits were already layered before this announcement. OpenAI's Codex pricing documentation describes rolling session windows and notes that weekly limits may also apply, and published guidance typically gives per-model ranges rather than a single fixed number. Consolidation therefore changes the shape of the constraint more than the existence of one. It also matters that a chat turn and an agentic coding run are not comparable units of effort: long context, high reasoning depth, attached files, plugin and tool definitions, and parallel subagents all multiply token use, which is exactly the kind of work the pooled budget now prices.
Plan-by-plan: who is exposed, and how much
List price does not change with this announcement, but price-per-usable-budget does, because one subscription now has to cover two product lines. Figures below are drawn from OpenAI's public help and pricing pages as of this writing, cross-checked against third-party trackers, and should be confirmed in the official checkout page before any procurement decision.
| Plan | Publicly listed price reference | Exposure to the pooled allowance | Most likely pinch point |
|---|---|---|---|
| Plus | US$20 per month for one user | Single person's chat and coding compete directly | A heavy afternoon of Codex work leaving too little for overnight drafting |
| Pro | Tiered at roughly US$100, US$200 and US$500 per month | Largest absolute pool, but also the most agent and long-context work | Frontier-model runs with big repos in context draining the pool early in a cycle |
| Business | Standard seats listed around US$25 per user monthly, or US$20 per user billed annually, with a two-seat minimum; higher-spec seats listed near US$100 to US$125 per user monthly | Seat already bundles ChatGPT, ChatGPT Work and Codex, so pooling aligns billing with how the seat is sold | Uneven burn across members and no per-seat view of what is chat versus code |
Two data points worth pinning to the wall for team planning, sourced to OpenAI's own help documentation: the Business overview page prices Standard seats at US$25 per user per month on monthly billing and US$20 per user per month billed annually, and states that the seat covers ChatGPT, ChatGPT Work and Codex; separately, the Codex pricing page says that Enterprise and Edu customers on flexible pricing face no fixed rate limits, with usage scaling to purchased credits. Read together, they sketch the escape valve for high-variance teams: buy flexibility rather than a fixed tier.
Scenario verdicts: who gains, who loses
Chat-first knowledge workers
Least affected. Because Instant mode is carved out of the pool, drafting, summarising, translating and quick questions keep running as before. Losses appear only when these users start handing longer reasoning tasks to higher-tier models, and then discover that the same budget is feeding their coding runs.
Developers running Codex daily
The clearest losers on flexibility, the clearest winners on transparency. Previously, an exhausted coding allowance did not stop chat work. Now both pull one lever, so schedule discipline matters more than raw quota. Watch the ratio rather than the total: if most burn is coding, a plan upgrade beats a top-up; if burn is spiky, credits are the cheaper hedge.
Agent and automation users
The highest risk category. Agentic sessions are token-hungry by construction, and the merged Chat and Work tabs push exactly this behaviour into the main interface, where a user can delegate a build, a slide deck or a web task and keep steering it mid-run. Anyone chaining long tasks should treat the shared pool as a schedulable resource, not a background allowance.
Business plan administrators
Predictability improves at the invoice level and worsens at the behaviour level. Seats already covered the two surfaces, so pooling removes double-counting confusion, but the effective cost per productive hour becomes harder to model while OpenAI's public documentation does not yet spell out per-plan token figures or a fixed cutover day inside December. Governance note: a policy on which work belongs on the pooled subscription and which belongs on metered API or credit-based usage is now a cost decision, not a tool preference.
A six-step checklist before the cutover
- Baseline the split. Log the last two weeks of chat versus coding activity, and note which model tier each job used; token weighting means plan choice should follow that mix, not the headline price.
- Open Settings and then Usage after the change lands, and screenshot the first two reset cycles to learn your real burn curve before committing to an upgrade.
- Route deliberately: keep routine turns on lighter models and reserve high-tier models for work where reasoning depth changes the output; complexity is now priced.
- Trim hidden load: large attached files, plugin and tool definitions and parallel subagent runs all inflate consumption per task.
- Move long jobs off-peak. Because one pool serves both surfaces, batch overnight is no longer free capacity that another product's clock would have reset.
- Decide the overage policy in advance, in this order: wait, upgrade, or purchase credits; teams should add a fourth branch, shifting burst capacity to metered or credit-based usage.
What remains open, and the risks to watch
OpenAI's guidance frames the change as starting in December rather than naming a single switch date, and public documentation stops short of plan-level token figures. Any workflow rewrite based on leaked or inferred numbers is premature. Three second-order effects deserve attention: first, cross-product cannibalisation, where a productive coding session quietly starves next-morning drafting; second, a behavioural nudge toward lighter models for cost reasons, which is a quality risk for work where a stronger model is the correct tool; and third, comparability, since a pooled allowance cannot be benchmarked against message-count limits, making vendor claims and internal reporting harder to align. The interface merge compounds this: with Chat and Work in one view, the quota becomes the only signal telling a user when to stop delegating and start prioritising.
FAQ
Does this mean I pay more?
No price change is stated alongside the pooling. The change is in what a plan buys: one allowance now covers ChatGPT and Codex, so value per dollar depends on your chat-to-code ratio.
Will ordinary chatting eat my coding quota?
Not on Instant mode, which OpenAI excludes from the shared allowance and describes as unlimited on current plans. Higher-tier model conversations are where chat begins to compete with code.
How do I see combined consumption?
In Settings, under Usage, where the merged remaining allowance and reset time will be displayed once the change is live.
What happens the moment the pool runs out?
Three documented options: wait for the reset, upgrade the plan, or buy additional credits to keep using eligible features. Teams with spiky demand can additionally look at flexible, credit-scaled arrangements.
Does Codex disappear into ChatGPT?
No. Codex retains a standalone experience inside the ChatGPT desktop app; what changes is the billing meter, not the workspace.
Are all plans affected?
OpenAI's Help Center names Plus, Pro and Business as the subscriptions moving to the shared allowance in December.
Bottom line for decision-makers: treat this as a resource-allocation event, not a repricing event. Users who barely touch Codex will feel nothing, while those who straddle conversation and code gain a single, legible budget and lose the ability to lean on two independent reset clocks. The right response is measurement, model routing and a written overage policy before December, so the pool is spent on the work that actually changes outcomes rather than on whichever task started first.