ChatGPT and Codex usage changes: how to prepare for December
If you use ChatGPT to work through a problem and Codex to build the solution, the connection between those tasks is about to matter when you check your remaining usage.
OpenAI has announced a shared allowance for ChatGPT and Codex, starting in December for Plus, Pro and Business. Here is what to understand before the transition, with a practical way to plan your work.
What OpenAI has announced
Agentic capabilities are features that can carry out steps towards a task, rather than only return a conversational answer. For your own planning, identify which tasks involve substantial investigation or execution and which need a straightforward response.
Tokens explain why message counts are not enough
Tokens are pieces of information a model reads or produces. The context can include instructions, files, conversation history and tool results, as well as the answer. OpenAI’s usage guidance explains why context, reasoning and tools can make superficially similar tasks use different amounts.
Consider these two example requests: “Explain this error message in three sentences” and “Investigate this error across my project, make a repair and run the relevant checks.” Each begins with one message. The second asks for a much broader piece of work.
That broader request may be the right choice. The useful question is whether the investigation is needed and whether the task has a clear stopping point. A smaller request that misses the actual problem is not automatically better.
What the shared allowance means for your working day
The planning implication is that activity in one product can affect the remaining allowance available for the other. Switching between ChatGPT and Codex is therefore not something to rely on for a fresh allocation.
Consider this example: you are improving a portfolio contact form. You need to understand a validation problem, implement a repair and explain the result in your project notes. Organise the work around those outcomes.
| Outcome | A bounded request | Review evidence |
|---|---|---|
| Understand the issue | Ask for a short explanation of the validation rule. | A clear description you can check against the form. |
| Repair the behaviour | Ask Codex for a focused change and relevant checks. | A reviewed change with evidence from the checks. |
| Record the result | Draft a concise project note from the verified change. | A note that names what changed and what was checked. |
The table is a workflow example, not a usage estimate. It does not assign a token cost to any step or guarantee which features your account can access. Check the available modes and current usage in your own account.
How to get useful work from your allowance
1. Define the result before the task starts
“Improve my website” leaves a lot for the assistant to decide. A clearer brief would be: “Investigate why the contact form accepts a blank email address. Propose the smallest suitable fix, check blank and valid inputs, and explain the result. Leave unrelated styling for a separate task.”
For product work, name the decision: “Compare these three onboarding ideas against the stated user problem and constraints. Separate evidence from assumptions. Finish with the questions I need to answer before prioritising.”
2. Give relevant context, then inspect the result
OpenAI recommends limiting source material and specifying the required output in its guidance on making usage last longer. Use that as a reason to select the right material, while retaining information needed for a sound answer.
For example, share the relevant form component, validation rules and error behaviour. Ask the assistant to identify missing context before making assumptions. After a repair, review the changed code and checks. The allowance does not replace your responsibility to understand what changed.
3. Match the model and mode to the work
Try a straightforward task with a suitable everyday option before reaching for the most demanding setting. For a difficult investigation, a more capable option may save repeated failed attempts. Judge the result as well as the usage.
The Instant exception is useful for deciding where simple conversation fits. It is not a reason to describe all ChatGPT use as unlimited: the announcement’s exemption concerns this shared allowance.
What is still worth checking
The announcement gives no exact December start day, per-plan token totals or task-to-allowance conversion. OpenAI expects most users will not face more frequent reset waits; that is an expectation, not an individual guarantee.
Before changing your subscription, identify the constraint you actually encounter. Note the mode, task and reset information shown. If you consider extra credits, check that they cover the feature you need. Do not buy more capacity solely because the measurement has changed.
Try a three-task review before December
Choose three familiar tasks: one explanation, one document or product task, and one coding task. Record the result you need, the mode you chose, what context you supplied and what required correction.
Use the exercise to find vague briefs, unnecessary context and work that needs stronger review. Then revisit it after the transition using your account’s actual usage display. Keep the focus on whether you finished worthwhile work and can stand behind the result.
Feature details checked on 3 November 2026. This article describes an announced December transition. Sources: OpenAI’s transition announcement
You May Also Like
Before you announce a feature: check who can actually use it
Scrum fundamentals: a practical guide with everyday examples
