From Shubhi K | Product & Market Analysis
ChatGPT and Codex Shared Limits: 3 Plans Move to One Allowance in December
On this page
From December, 3 ChatGPT plans stop treating chat and coding as separate budgets. OpenAI says the usage allowance on Plus, Pro and Business will be shared across ChatGPT and Codex, counted in tokens. ChatGPT Codex shared limits change who runs out first, and OpenAI has not printed a token total for any of the 3 plans.
Key takeaways
- Plus, Pro and Business move to one shared allowance in December. OpenAI's Help Center says the allowance on those 3 plans will be shared across ChatGPT and Codex, with usage counted in tokens.
- Instant chat is exempt, and that exemption decides most outcomes. OpenAI says Instant chat will not draw from the allowance, so a user who never picks a reasoning model or Codex should see no change.
- The only published per-plan figures are wide Codex estimates. OpenAI estimates 15 to 160 local GPT-6.1 Sol messages per 5 hours on Plus, a 10.7x spread, and calls them estimates, not limits.
- Heavy Codex users are the exposed group. A developer who already reaches the Codex window will now draw reasoning-model chat from the same pool, and OpenAI has not said how big that pool is.
Short answer. Starting in December, Plus, Pro and Business subscribers spend one token-based allowance on both ChatGPT and Codex. Instant chats are exempt. OpenAI expects most users not to wait for resets more often than now, but it has not published the allowance size, so heavy Codex users should plan for less headroom.
What changes in December for ChatGPT and Codex shared limits
The change is one sentence in an OpenAI Help Center article. As quoted by DataNorth, it reads: "Starting in December, Plus, Pro, and Business plans' usage allowance will be shared across ChatGPT and Codex." Notebookcheck and ARY News carried the same wording a day later.
OpenAI frames the move as part of bringing "the agentic capabilities of Work into the main ChatGPT experience", per Notebookcheck's report. That framing matters. The company is not describing a price cut or a price rise. It is describing a merger of two meters into one.
The Help Center defines tokens as "the units of information a model processes and generates", according to DataNorth's quotation. Tokens are the same unit OpenAI's API bills in. Counting a subscription in tokens means a long reasoning answer and a short one no longer cost the same.
That is a real shift in how a subscriber should think. A message count is easy to reason about. A token budget depends on the model you pick, the length of the context, and how much work the model does before replying.
Instant chat sits outside the shared pool
The most important line in the announcement is the exemption. OpenAI says "Instant chat will not draw from your usage," a sentence quoted identically by Notebookcheck and ARY News. A subscriber who uses ChatGPT only in its default fast mode should notice nothing.
I think this exemption is the whole design. It protects the largest group of paying users from the change, and concentrates the effect on the minority who run reasoning models and coding agents.
OpenAI says December, not a specific day. Notebookcheck and ARY News say the help article was updated on 2 November. DataNorth, publishing on 2 November, says the article appeared at the end of October. The month is consistent across all 3 reports, and that is the only date you should plan on.
ChatGPT usage allowance and Codex limits before December
To see what changes, start with what OpenAI already prints. The Codex pricing page is more specific than the Help Center, and it already describes a partial merger.
The page states: "ChatGPT Work and Codex share usage. ChatGPT Work usage inside ChatGPT uses the same pricing, credits, and usage limits as Codex." So the agentic Work features and Codex already draw from one pool. December extends that pool to the rest of ChatGPT, minus Instant.
The same page adds that "Usage limits are shared with other agentic features once pricing for those features is effective," and names ChatGPT for Excel on Plus and Pro as the current example. The direction of travel was visible before the announcement.
| Plan | Price as listed | Five-hour window | Weekly limit | Extra usage |
|---|---|---|---|---|
| Plus | $20 per month | Yes, with per-model estimates | "May also apply", no number | Credits, price depends on plan |
| Pro | $100, $200 or $500 per month | "No five-hour limit" currently | "May also apply", no number | Credits, price depends on plan |
| Standard Business | $20 per seat annual, $25 monthly | Yes, same estimates as Plus | Not stated per seat | Workspace credits with flexible pricing |
| Enterprise and Edu, flexible pricing | Contact sales | "No fixed rate limits" | Scales with credits | Workspace credits |
All cells are OpenAI's own wording or figures from the Codex pricing page. "No number" means OpenAI prints no figure, not that no limit exists. The page shows no last-updated date.
The Plus estimates are ranges, not caps
For Plus and Standard Business, OpenAI publishes local messages per 5 hours by model. GPT-6.1 Sol runs from 15 to 160, GPT-6 Astra from 5 to 45, and GPT-6 Luna from 350 to 3,000. OpenAI says plainly: "These estimates are not fixed message limits."
The width of those ranges is the first finding of this post. Dividing the high end by the low end gives a spread of 8.1x to 10.7x depending on the model. A planning number with a 10x spread is not a planning number.
How one shared allowance drains under heavy coding use
Two separate meters fail independently. If your Codex window runs dry, your reasoning chats still work, and the reverse is also true. One shared meter removes that independence.
The mechanism is simple arithmetic on a pool. Every token spent in Codex is a token not available for a reasoning chat later in the same window. The heaviest workload sets the pace for everything else.
Cloud tasks and speed modes burn faster
"Cloud tasks may use more of your allowance than local messages," OpenAI already warns. The pricing page also lists speed modes that draw more of the included usage. Fast mode uses 2.5x included usage, and Ultrafast uses 8x for GPT-6 Astra and GPT-6.1 Sol.
Apply those multipliers to a fixed pool and the effect is direct. A window that holds 40 standard tasks holds 16 in Fast mode and 5 in Ultrafast. Ultrafast is listed for Pro $500, so this mostly hits the top tier, which is also the tier most likely to run long agent sessions.
Coding agents read files, run tests and re-read output. Each step adds tokens to the context the model processes. A token-counted pool therefore charges more for exactly the agentic, multi-step work Codex is sold to do.
The API side of this is covered in the analysis of why agent token costs vary so widely between runs. Inside a subscription the variance does not show up on an invoice. It shows up as an earlier reset wait.
Who ChatGPT Codex shared limits actually affect
The question most coverage skips is who notices. OpenAI's own forecast is that "most users won't need to wait for their usage to reset more often than they do now," as quoted by DataNorth and ARY News. I believe that forecast for most users. It is the tail I would worry about.
| User profile | What they use | Likely effect | Confidence |
|---|---|---|---|
| Chat-only subscriber | Instant chat | None. Instant does not draw from the allowance. | High, stated by OpenAI |
| Reasoning-chat user, no Codex | Thinking or reasoning models in chat | Now metered in tokens. Effect depends on a pool size OpenAI has not disclosed. | Low |
| Light Codex user | Occasional local tasks plus chat | Small. Combined use likely stays under one window. | Medium |
| Heavy Codex user on Plus | Daily agent sessions near the 5-hour estimate | Reasoning chat now competes with coding inside the same window. | Medium |
| Pro user on long cloud tasks | Cloud tasks, Fast or Ultrafast | No 5-hour window today, so the weekly limit is the binding constraint. | Low, weekly figure unpublished |
This is a model built on OpenAI's published rules, not on usage data. Zan Digital has no first-party telemetry on ChatGPT or Codex consumption. Confidence reflects how much of each row rests on a disclosed figure.
If your team uses ChatGPT for drafting, summarising and quick questions in Instant mode, the change does not reach you. That covers a large share of paying seats in most firms I have seen, though OpenAI publishes no split.
The developer who already hits the window
This is the exposed profile. A Plus developer who runs into the Codex 5-hour estimate today has already spent the shared Work and Codex pool. From December, any reasoning chat they open in the same window draws from that same exhausted pool.
My view is that this user experiences the change as a cut, even if the total allowance grows. Perception follows the moment of refusal, not the size of the pool.
Pro has "no five-hour limit" today, per the pricing page, but "Weekly limits may also apply." A Pro user cannot run out within a window, only within a week. For that user a shared pool converts heavy chat days into fewer coding days later in the week. The published history of Pro tier changes is tracked in the dated timeline of every 2026 ChatGPT plan change.
What OpenAI has not disclosed about AI plan limits
The announcement is specific about structure and silent on size. That combination is common across AI subscriptions, and it is the pattern examined in the proposed standard for rate limit transparency.
No token figure appears for any plan. Without it, nobody outside OpenAI can say whether the merged pool is larger than the old Codex pool, the same, or smaller. The statement that most users will not wait more often is a forecast about behaviour, not a disclosure about capacity.
The weekly limit
Both Plus and Pro carry the phrase "Weekly limits may also apply" with no number. For Pro, which has no 5-hour window, that unpublished weekly figure is effectively the whole limit. I would treat it as the single most important missing number.
The Codex page lists credits per million tokens by model, so a heavier model costs more credits. It does not say whether the included allowance applies the same weights. If it does, model choice becomes a budget decision for every user, not only developers. The confusion this creates is examined in the post on why credits and tokens confuse AI buyers.
ChatGPT Business limits and the per-seat question
Business is listed at "$20/ user / month" with a footnote: 2 or more users, billed annually, and "$25 per user per month when billed monthly." That is a 25% premium for monthly billing. A 10-seat team pays $2,400 a year on annual terms or $3,000 on monthly terms.
On usage, Standard Business carries the same Codex estimates as Plus. So a Business seat and a Plus subscription buy the same published coding allowance at the same annual price. What Business adds is workspace administration and, with flexible pricing, the option to buy workspace credits.
Allowances are per seat, not pooled
Nothing on the pages reviewed suggests a team-wide pool. Each seat appears to carry its own allowance. That means one heavy developer cannot borrow headroom from five colleagues who only use Instant chat, unless the workspace buys credits.
That is the practical problem for an operations lead. You will have seats that never touch the pool sitting next to seats that empty it by lunchtime. Seat-based pricing has the same flaw elsewhere, as the analysis of seat compression in SaaS pricing sets out.
The December wording names Plus, Pro and Business only. The Codex page says Enterprise and Edu plans "without flexible pricing have the same per-seat usage limits as Plus for most features". Plans with flexible pricing "have no fixed rate limits", it says. Whether the shared pool reaches them is not stated. Do not assume either way.
Where this analysis of the shared allowance is weakest
This post models an allowance whose size is unknown. That is its central weakness, and it limits every conclusion above to direction rather than magnitude.
OpenAI's Help Center refused automated retrieval during research for this post. Every quotation about December comes through DataNorth, Notebookcheck and ARY News. Their wording matches each other, which is reassuring, but 3 restatements of one page are still one source. The Codex pricing figures, by contrast, were read on OpenAI's own page.
The merged pool could be bigger
The strongest counter-case is simple. OpenAI could size the shared pool well above the old Codex pool, so heavy coders gain headroom on days they do little chat. A shared pool also lets a quiet coding day subsidise a heavy research day, which separate meters never allowed. If that happens, the change is a net gain for most mixed users, and my framing of it as a risk would be wrong.
The 5-hour figures are labelled estimates and the page shows no update date. Third-party trackers report that the 5-hour window was removed and reinstated for some plans during 2026. Any number in this post could be stale within weeks, which is why each one carries its retrieval date.
Preparing a team for one AI usage budget
You cannot plan against a number OpenAI has not printed. You can plan against your own usage pattern, which is the one dataset you control.
Ask each seat holder one question: do you use reasoning models or Codex daily? Instant-only users can be ignored for this change. The remaining group is your exposure, and in most teams it will be a short list of developers and analysts.
Reports of the Help Center article say combined remaining usage and reset times will appear under Settings, then Usage. The Codex page already points users to a usage dashboard. Have exposed users screenshot it at the end of each working day in the first week of December. That gives you a baseline OpenAI will not.
Decide in advance who gets credits
Plus and Pro users who reach their limit "can purchase additional credits", OpenAI says. It adds that "Credit purchase prices and applicable discounts depend on your plan or agreement." Set a monthly credit ceiling per exposed user before December. A ceiling decided in advance is a budget. One decided at the moment of refusal is an expense. The broader trend of plans quietly delivering less is covered in the analysis of AI shrinkflation in usage limits.
Frequently asked questions
Will ChatGPT and Codex share usage limits?
Yes. OpenAI's Help Center says that starting in December, the usage allowance on Plus, Pro and Business plans will be shared across ChatGPT and Codex. Usage will be counted in tokens. Instant chat will not draw from the allowance. OpenAI has not named a specific start date or published the size of the shared allowance for any plan.
Does ChatGPT Instant count toward the Codex limit?
No. OpenAI states that Instant chat will not draw from your usage under the shared allowance. Only other ChatGPT use, such as reasoning models and agentic Work features, plus Codex local and cloud tasks, counts against the pool. A subscriber who only uses Instant chat should see no change in December.
How many Codex messages do you get with ChatGPT Plus?
OpenAI publishes estimates, not caps. For Plus it estimates 15 to 160 local GPT-6.1 Sol messages per 5 hours, 5 to 45 for GPT-6 Astra and 350 to 3,000 for GPT-6 Luna. It states these are not fixed message limits, that cloud tasks may use more allowance, and that weekly limits may also apply.
Does ChatGPT Pro have a usage limit for Codex?
OpenAI's Codex pricing page says Pro plans currently have no five-hour limit, but weekly limits may also apply. Pro is listed at $100, $200 or $500 per month, with Ultrafast mode on the $500 plan. OpenAI does not publish the weekly figure or a usage multiple over Plus for each Pro tier on that page.
Are ChatGPT Enterprise and Edu affected by the shared limit?
OpenAI's December wording names only Plus, Pro and Business. Its Codex pricing page says Enterprise and Edu plans without flexible pricing have the same per-seat limits as Plus for most features. Those with flexible pricing have no fixed rate limits and scale with credits. Whether the shared pool applies to them is not stated.
What happens when you run out of ChatGPT and Codex usage?
You wait for the reset, upgrade, or buy credits. OpenAI says Plus and Pro users who reach their limit can purchase additional credits without upgrading, and that credit prices depend on the plan or agreement. Business, Edu and Enterprise workspaces with flexible pricing can buy workspace credits. Reports say remaining usage and reset times will show under Settings, then Usage.
Where to start before the December switch
This week, list every seat holder who uses Codex or a reasoning model at least daily. For most teams that list is under a quarter of seats, and it is the only group worth planning for.
Then write one number next to each name: the monthly credit spend you will approve if that person hits the shared limit. When December arrives, you will be choosing from a decision already made rather than negotiating with a reset timer.
Related analysis
For what heavy agent use costs outside a subscription, see the daily cost of terminal coding agents. For how other vendors word their limits, see the fair use clauses inside unlimited AI plans.
References
- OpenAI, Codex pricing (redirected from developers.openai.com/codex/pricing), retrieved 12 November 2026. Used for per-model 5-hour estimates, Pro and Business wording, plan prices, speed-mode multipliers, credits and Enterprise and Edu terms.
- OpenAI Help Center, Understanding ChatGPT and Codex usage limits, updated late October or 2 November 2026. Not retrievable by automated fetch; quoted via the three reports below.
- DataNorth, ChatGPT and Codex will share one usage limit in December, 2 November 2026. Used for the verbatim shared-allowance sentence, token definition and reset forecast.
- Notebookcheck, ChatGPT and Codex will share one usage limit starting in December, 3 November 2026. Used for the Instant exemption and Work framing quotes.
- ARY News, OpenAI to combine ChatGPT and Codex usage limits for premium subscribers, 3 November 2026. Used to corroborate quotations and the Settings, then Usage location.
Weakest point in the source base: the December announcement could not be read on OpenAI's own page and rests on three secondary reports quoting it identically. Figures last verified 12 November 2026. Re-check the Codex pricing page after the December switch.
Related reading