OpenAI has announced that starting in December 2026, ChatGPT and Codex will share a single usage allowance. The change applies to the Plus, Pro and Business plans, and chats in Instant mode will not draw from the shared allowance. For users who combine conversations in ChatGPT with coding work in Codex, both activities will now affect the same remaining balance. However, how much allowance is consumed is not determined by the number of requests alone. It also varies with the amount of information processed, the model used and the settings applied. Even under the current system, as of November 1 (Japan time), usage included in a monthly plan and purchased credits are consumed under different conditions. Understanding these differences is essential to making sense of the changes coming in December.
ChatGPT and Codex Allowances to Be Combined, With Instant Excluded
The change is part of a broader move to integrate Work's agent features into ChatGPT itself. As regular conversation and the ability to hand tasks to an AI agent continue to converge, new guidance published by OpenAI sets out a plan to share usage allowances across ChatGPT and Codex on Plus, Pro and Business. Chats in Instant mode are excluded from the shared allowance, but that does not mean every mode can be used without limit.
The idea of sharing allowances across multiple features has already been put in place. According to the current Codex usage guide, Excel and Workspace Agents, in addition to the available Work and Codex features, already share allowances and credits. Regular Chat, on the other hand, is not included in the current display of remaining Work/Codex credits. That describes what the balance display covers, though, and does not mean every feature has its own independent usage cap. From December, users will be able to see a combined balance covering both ChatGPT and Codex usage.
The new guidance says the "Usage" section in settings will show the remaining shared allowance and the reset time. However, the specific start date in December, the amount available on each plan, the reset cycle and the consumption conditions by feature and model have not been disclosed. The explanation of the covered plans also does not make clear how Free, Go, Enterprise and Edu will be handled going forward.
Why Request Counts Alone Don't Explain Token Consumption
A request to answer a short question and a request to read through a large set of documents and keep working on a complex task are each just "one request," yet the amount of information the AI processes can differ enormously. A token is the unit of information an AI model uses when processing or generating text and other content. Sharing allowances between ChatGPT and Codex concerns which features' usage is added together, and it should be considered separately from the number of tokens processed in any given task.
According to OpenAI's current pricing documentation, the tokens processed include not only the prompt a user types but also attached files and conversation history. Results returned by tools and the responses the model generates also use tokens. As a result, even if the text a user enters is short, a task that references a large volume of material may process far more information than the on-screen prompt suggests.
For example, asking about a short function and repeatedly revising code while examining multiple source files cannot reasonably be compared as the same "one request." In the latter case, the information processed may grow each time a new file is loaded or a tool's output is received during the work. In other words, what drives consumption is less the length of the prompt itself than the information actually referenced and how the work proceeds.
The current Codex usage guide also explains that long-running tasks can consume significantly more of the allowance than short requests. Consumption conditions further depend on the model used and reasoning settings. Because whether the task runs locally or in the cloud, the processing speed setting and the tools used all play a role, it is difficult to predict how fast the balance will drop from the number of requests or the number of lines of generated code alone.
The message counts listed in the pricing documentation are likewise not fixed numbers guaranteed to be available. Plus and Standard Business currently list estimated usage per five-hour window, while Pro currently has no five-hour limit, although weekly limits may apply. These all reflect the current system as of November 1 and do not represent the terms of the shared allowance beginning in December.
When comparing actual consumption, it is necessary to consider not just the number of requests but also the volume of material processed, the model used and various settings. Between a usage pattern of many short conversations and one in which an AI agent is handed long-running work, the one with more requests does not necessarily consume more of the allowance.
How Tokens Differ From Usage Allowances and Credits
Under the current system, choosing "Fast" for processing speed results in different consumption multipliers depending on whether you are using the allowance included in your monthly plan or purchased credits and similar balances. Based on OpenAI's pricing documentation, the conditions as of November 1 compare as follows. In both cases, the figures assume the Standard speed of the same corresponding model as 1x, and treatment varies depending on the model and the terms under which it is offered.
| Speed setting | Consumption multiplier for plan usage | Consumption multiplier for purchased credits, etc. |
|---|---|---|
| Standard | 1x (baseline) | 1x (baseline) |
| Fast | 2.5x | 2x |
These figures show the multipliers for consuming allowances or credits, not how many times faster actual processing becomes.
With Fast as currently offered, the consumption multiplier is 2.5x for plan usage, while it is 2x when purchased credits and similar balances are used. This difference is a separate matter from the capacity of the shared allowance being introduced in December. When choosing a processing speed, it is therefore necessary to distinguish whether you are consuming the allowance included in your monthly plan or using credits you bought separately.
Credits are also a billing unit for additional use of eligible features, not the tokens the AI processes. OpenAI's pricing documentation likewise explains that the price of credits alone does not determine how much usage is included in a monthly plan. For that reason, the total size of the shared allowance from December or the number of questions available cannot be worked backward from the multipliers above. The capacity of the allowance, the amount of information processed in an individual task and the applicable consumption multiplier need to be checked as separate measures.
When You Hit the Limit: Buy Credits or Wait for a Reset
For the new system starting in December, OpenAI lists three options when a usage limit is reached: waiting for the next reset, changing to a higher plan, or purchasing credits usable on eligible features. However, buying credits does not mean every mode can be used without condition. The features for which additional use is permitted and the applicable conditions need to be checked.
Under the credit system for personal plans as of November 1, Plus and Pro users first consume the allowance included in their monthly plan. After reaching that limit, they use credits they have already purchased on features that support additional usage. Which features are available varies with conditions such as the account. Purchasing credits secures a balance for additional use of eligible features and is not an operation that restores the monthly plan's allowance itself.
Separately, the current system also offers a paid option to restore allowances, called "instant reset." It applies to individual Plus and Pro accounts, and whether it can be purchased differs by account. Business and other plans are not covered, and there is no guarantee it will be available under the same conditions after December.
Applying an instant reset immediately restores the five-hour and weekly allowances applicable to that account. However, the seven days until the next weekly reset are not counted from the moment of purchase. They start from the time the first request is sent to Work or Codex after the instant reset is applied.
When paying an additional fee, it is therefore important to confirm not only the amount but also what is being purchased. Whether you continue using eligible features with additional credits or restore your in-plan allowance with an instant reset changes the remaining balance and the terms of use after purchase. If you plan your work around the weekly reset time, when the period after an instant reset begins can also factor into the decision.
OpenAI expects that, with this consolidation, most users will not have to wait for resets more often than they do now. That is not, however, a guarantee of how much any individual user can use. For those who combine everyday ChatGPT conversations with long-running development work in Codex, it will matter which tasks consume how much of the shared allowance. From December, comparing the remaining balance and reset time shown under "Usage" in settings with your actual workload should make it easier to decide whether to wait for the allowance to recover, change the model or speed setting, or buy additional credits.