Playbooks · Coding tasks
Baseline usage before the shared allowance
Written November 2, 2026
From December, ChatGPT and Codex share one token allowance on Plus, Pro, and Business. Measure your November split now — OpenAI publishes no sizes.
What is changing
OpenAI's Help Center article Understanding ChatGPT and Codex usage limits states the change in three sentences. First: "Starting in December, Plus, Pro, and Business plans' usage allowance will be shared across ChatGPT and Codex as we bring the agentic capabilities of Work into the main ChatGPT experience." Second: "Your usage for ChatGPT and Codex will be measured in tokens, the units of information a model processes and generates. How much of your allowance you use will depend on the model you choose and the task you ask it to do. More advanced models and complex tasks generally use more of your allowance. Instant chat will not draw from your usage." Third: "We expect most users won't need to wait for their usage to reset more often than they do now. You will be able to see your combined remaining usage and reset times in Settings → Usage. If you reach your limit, you can wait for it to reset, upgrade your plan, or buy credits to continue using eligible features."
Three things about that wording matter for planning. The page names no exact day in December, so do not plan around 1 December as a start date. It publishes no allowance size for any plan, in tokens or in any other unit, so there is no number to compare a November week against. And the sentence about most users waiting no more often is a forecast — "we expect" — not a guarantee about your usage. What the page does give you is the shape of the change: one allowance, measured in tokens, with Instant chat outside it, visible in one place, with three options at the limit. The rest of this page is about measuring your own use against that shape before it arrives, because your own November numbers are the only before picture that will exist.
One earlier statement needs placing, because it still circulates. On 28 October, OpenAI's Codex lead posted on X that bringing Chat and Work together was "not rolling out just yet", and that the team would proceed carefully. That post was about the Chat and Work experiences coming together, not about usage limits. The Help Center page quoted above is the current statement of the December allowance change. Treat the page, not the earlier post, as the position to plan from — and note that the page itself gives no day and no sizes, so a plan built on either is still a guess.
What already shares an allowance
The December change is not the first sharing. Today, Codex, ChatGPT Work, ChatGPT for Excel and Workspace Agents already use one shared allowance and credit pool, where those features are available on your plan, according to OpenAI's Codex help article. OpenAI's Codex pricing page states the same for Work and Codex: "ChatGPT Work and Codex share usage." What December adds is ordinary ChatGPT chat, on the advanced models, joining that same allowance. Instant chat stays outside it. Auto-review safety checks in Codex are also outside it: when signed in with a ChatGPT account, they are free and do not count toward plan usage limits, per the Codex help article. Codex used with your own API key is billed at API rates instead, and is not part of the plan allowance at all.
This is a different vendor's allowance from Anthropic's. Bringing your own Anthropic API key to an agent, covered in the page linked at the end, pays OpenAI nothing and draws on no ChatGPT allowance — it bills through Anthropic at API rates. Do not mix the two up when comparing costs: the December sharing affects ChatGPT and Codex on a ChatGPT plan, and nothing else.
Record your November split
Record usage once a week through November, in the same format each time, so December has something to compare against. OpenAI gives no sizes, so the comparison that matters is your own: how much of the allowance a chat-heavy week takes versus a Codex-heavy week, and how close to the limit each kind of week gets you.
- Check Settings → Usage in ChatGPT. Open Settings from your profile menu and select Usage. Note the remaining usage and the reset times shown, and the date. This is where the combined remaining usage and reset times will appear after the change. Write down what it shows today, in your own words, with the date — a screenshot works too, with anything sensitive removed.
- Check Codex separately. In an active Codex CLI session, enter
/statusto see remaining usage and reset times for the Codex side. In the Codex app, open Usage & Billing. Note the same figures, with the same date. The two checks together are your split: what ChatGPT shows and what Codex shows, on the same day. - Label the week. Beside each week's figures, write one line about what the week was: mostly chat, mostly Codex, or mixed, and which models you used most. A week of routine Codex work on a light model and a week of heavy model work are not the same week, and the Help Center page says model and task decide how much allowance a request uses. Without the label, the numbers cannot be compared.
- Note when you hit a limit, if you do. If a limit banner appears, note the date, which surface you were in, and the reset time shown. That is the only record of how often waiting for a reset happens today — the baseline for judging, in December, whether it happens more often.
- Keep the log somewhere you will find in December. One note, four or five weekly entries, is enough. The log's job is narrow: in December, when chat and Codex draw on one allowance, you can look back and see whether a week like this one used to fit comfortably, or used to run close to the limit.
Make the allowance last
These habits use only settings and choices OpenAI documents. None of them needs code, and none of them changes what the allowance is — they change how fast you spend it.
- Use Instant chat for everyday questions. Instant chat is the one thing the Help Center page says will not draw from the shared allowance. Use it for quick lookups, rewording, translations and simple explanations. Save the bigger models for work that needs them, and more of the allowance stays available for Codex.
- Pick a lighter model for routine Codex work. The pricing page's estimates for local messages per five-hour period, for Plus and Standard Business, show how much model choice matters: GPT-6 Astra is estimated at 5–45 messages, GPT-6.1 Sol at 15–160, GPT-6.1 Terra at 35–285, and GPT-6 Luna at 350–3,000. The page states these are estimates, not fixed message limits, and that cloud tasks may use more. The pattern is still clear: routine work on Luna or Terra leaves far more room than the same work on Astra. Use the heavier model where a wrong step is expensive, and the lighter one where a checkable result makes a retry cheap.
- Leave Fast mode off unless you need the speed. Fast mode uses included subscription usage at 2.5 times the Standard rate, per the pricing page. If you are not waiting on the answer, Standard gets the same work for less of the allowance. Ultrafast, where available, uses it faster still.
- Start fresh chats instead of endless ones. A long chat carries its whole history into every new turn, and the pricing page states that prompt length alone is not a reliable estimate of cost because model choice, context, reasoning, tool use, retrieval and caching all change it. When a task is done, start a new chat with a short summary of the goal and what is finished. A new chat does not reset your limits; it stops the history from being re-sent.
- Keep prompts tight. Give the agent the files it needs and a clear definition of done, and tell it to stop and ask before anything large. A vague request that runs for a long time uses substantially more than a short, specific one — that wording is from the Codex help article, and it is the cheapest habit on this list.
When you reach the limit
The Help Center page lists three options: wait for the reset, upgrade the plan, or buy credits to continue using eligible features. Check the reset time in Settings → Usage before choosing: waiting is free, and a reset hours away is a different decision from a reset days away. Do not buy credits in bulk before December on the strength of this change alone. The page states no credit prices for December overflow, and credits bought now are spent under today's rules, not December's. Wait until the shared allowance is live and your November log tells you how a real shared week compares with a November week, then decide whether the limit you hit is a weekly pattern worth paying to remove, or a single heavy week that waiting would have covered.
What can go wrong
- Planning around a start date OpenAI has not given. "Starting in December" is a month, not a day. A baseline recorded in late November is still a baseline whenever in December the change lands; a plan that assumes 1 December is not.
- Comparing November to December without labels. An unlabelled usage figure cannot tell you whether December got worse. The weekly label — chat-heavy, Codex-heavy, which models — is what makes the comparison possible.
- Treating the "most users" sentence as a promise. It is OpenAI's expectation, stated on the page. Your November log is the evidence for your own usage; if December brings more waiting, the log, not the sentence, is what you argue from.
- Assuming image, voice and upload limits join the pool. The Help Center page does not say how image generation, file uploads, or regular chat voice fit into the shared allowance. Do not assume they are in or out; check the page again when the change lands.
- Mixing vendors' allowances. Anthropic-key billing and API-key Codex use are separate from the ChatGPT plan allowance. A cost comparison that blends them will misstate both.
Limits
- The December sentence on the Help Center page names Plus, Pro and Business only. It does not mention Free, Go, Enterprise or Edu plans, and this page makes no claim about them either way.
- No allowance size, exact start day, credit price, or per-model cap treatment is stated on the page, and none is stated here. Any figure of those kinds found elsewhere is not OpenAI's published position on this page.
- Usage figures on the pricing page are estimates for local messages in a five-hour period, not fixed limits, and cloud tasks may use more. The baseline method on this page records what your own account shows, not what any estimate predicts.
Related playbooks
- Qwen Code 0.28 runs one shared agent loop
- Stay inside the Muse Code subscription allowance
- ChatGPT and Codex share one token budget
- Add tests to an untested module with an agent
Sources
- https://help.openai.com/en/articles/20001625-understanding-chatgpt-and-codex-usage-limits
- https://help.openai.com/en/articles/11369540-using-codex-with-your-chatgpt-plan
- https://learn.chatgpt.com/docs/pricing
Back to: Coding playbooks hub · Also: Pick the cheapest model for the job · Stop an agent running up a bill · Use your own Anthropic API key with an agent: what it costs, and the billing trap to check first