For a long time, our main concern while opening ChatGPT was getting a good response. Now there is another question that starts with those who use it more than just talking: How much margin is left before a limit is encountered. And not because all users are exhausting their quota, but because such tools have ceased to be simple places to ask questions and have become platforms for analyzing, creating and executing complete tasks. We’ve gained potential, yes, but we’ve also had to learn a language of tokens and assignments that were far from our conversations before.
Now OpenAI has announced a concrete change in the way its tools are used. In an email we received, the company explains that in the coming weeks it will begin testing an experience that combines chat and work, initially with a small group and with the goal of expanding later. Another decision included in the announcement: from December 2026, ChatGPT and Codex will share usage allocations on Plus, Pro and Business plans. Of course, the integration of the two tabs doesn’t mean the disappearance of Codex as an independent experience within the desktop application.
Let’s take an example to understand the difference. We can open ChatGPT, upload a document, and ask it to analyze it, but use Work to assign it a broad task that involves consulting files, using applications, and executing various steps. If we need to build software or modify a programming project, Codex comes into play, specifically specialized in those tasks. These are different experiences, although they don’t always differ in what they allow you to do. Until now, that difference was also reflected in their limits: Work and Codex shared a mapping, while conventional chat models had their own limits.
The same goes for working with AI
That separation offers users some margin. In the middle of a task, ChatGPT might suggest that we continue working, but we had the option of accepting the offer or staying in the conversation and using our own limits on chat rather than resorting to shared assignments of work and codex. We did not always achieve the same capabilities, although for some tasks the results may be quite similar. With the announced changes, this possibility of saving the work quota loses part of its effectiveness: the operations performed with the models included in the new allocation will also be excluded from it, even if we decide to continue in the chat.
The key is how those costs are discounted. According to OpenAI, not all requests will have the same weight: more advanced models and complex operations usually take up a larger portion of the available allocation. The company talks about tokens, the units of data that its models process and generate, but has not yet detailed how the new system will translate into usability for each plan or how much specific tasks will use. Therefore, we cannot establish a reliable equivalence in reports, conversations or projects. Once the quota is over, you have to wait for it to recover or resort to available payment options.
There is one word that stands out in particular in the statement: “Limitless.” The company ensures that the recipient’s plan includes immediate without taking new allocations, although its own documentation establishes restrictions against abuse and maintains specific limits for certain functions. Nor does it establish a general and universal threshold that allows knowing when these protections can be applied. And there is another important problem: instantiation can solve many everyday needs, but it does not necessarily provide the same capabilities as advanced reasoning models or self-working tools. For those who use ChatGPT as a professional tool, this difference is far from a minor detail.
There is another statement in the email that deserves special attention. OpenAI predicts that most users won’t have to wait that often to reclaim their allocation, though it doesn’t explain how it came to that conclusion. It does not even provide statistics that allow comparison of current conditions with future ones. It’s true that on October 29 it announced an increase in Work and Codex quotas for some Pro and Business Premium plans, but that didn’t include the equivalent improvements to Plus. So, for now, we have a reassuring forecast from the company and an unanswered question: How much will the experience change for those who pay to use these tools.
This transformation does not come out of nowhere. What started as a free chatbot in 2022 has grown to include memberships like Plus, Team and Pro, until it has grown into a platform capable of performing tasks beyond answering questions. The economies of scale of the business have also changed: In May, OpenAI president Greg Brockman said the company expected to spend $50 billion on computing by 2026, according to Reuters. This is a figure that helps understand the costs of operating these systems, although it does not demonstrate that the new limit policy directly responds to those costs.
The definitive proof will not come with the announcement, but when customers start working on the new terms. OpenAI indicates that you can check the available allocation and its reset date in Settings → Usage; If it runs out, the options are to wait, change plans or purchase credits for supported features These are concrete options, but the question they don’t answer remains open: how much better work each subscription will allow than today. Until we know those figures and see their impact, integration will be a promise of benefits whose consequences have yet to be measured.
Picture | Xataka with Nano Banana
In Xataka | Google has improved its search engine so much with AI that it faces a problem: it has to make it worse to make money.
