OpenAI is merging the existing Chat and Work areas in ChatGPT while making ChatGPT and Codex draw from a shared token allowance. This mainly affects people who use ordinary conversations alongside longer tasks, file-based work, or coding jobs. At the same time, new device restrictions make switching between multiple paid accounts harder, while additional Codex integrations could increase consumption further.
One interface for chats and assigned work
The ChatGPT desktop app currently separates ChatGPT from Codex and divides ChatGPT itself into Chat and Work tabs. Codex is an assistant focused on programming tasks, while Work covers larger jobs that may be carried out with some independence. This arrangement can become confusing when you no longer know which area contains a project, file, or assigned task.
According to an OpenAI email sent to selected users, Chat and Work will therefore be combined in one interface. The company reportedly plans to test the experience with a limited group before rolling it out more widely. Codex, however, will remain a separate experience inside the desktop app.
In practical terms, you should be able to discuss an idea in a chat and have it carried out in the same place. OpenAI’s examples include building a website, creating a presentation from uploaded files, and completing tasks on the web while you review and guide the work. This makes the boundary between receiving a single answer and assigning a multistep job less visible.
The approach is not entirely new. According to the reports, Anthropic had already combined Claude’s chat and Cowork functions in one screen in September. The ChatGPT change is still significant because its simpler interface will be tied to a new way of counting usage.
Shared tokens change how usage works
A token is a small unit of information that a language model processes while reading an input and producing an answer. Token usage is not simply the number of messages you send: long documents, more complex models, and multistep tasks can consume substantially more capacity than a short question. According to a summary of OpenAI’s updated rules, Plus, Pro, and Business subscriptions will receive a shared ChatGPT and Codex allowance starting in December.
Two types of usage that previously felt more separate will therefore compete for the same supply. If you use ChatGPT to create a presentation from several files in the morning and give Codex a substantial programming job later, both will count against the same allowance. OpenAI says, according to TechCrunch, that the change should not affect most people, although some may need to wait for resets more often depending on their model choices and usage.
Your remaining allowance and the time of the next reset are expected to appear in Settings. When the allowance is exhausted, the report identifies three options: wait for the reset, change plans, or buy additional credits. The available sources do not state the exact allowance sizes or prices for extra credits.
Basic immediate chats using so-called Instant models are not expected to count against the shared allowance. An Instant model is a model variant designed to provide quick responses. ChatGPT may therefore remain broadly available, even if more demanding functions pause sooner than ordinary conversations.
The actual impact is likely to depend heavily on your workflow. A sponsored article about the cost of AI coding agents says a single agentic programming task can use 10 to 50 times as many tokens as a basic chat prompt. Agentic means that the system plans, generates code, runs tests, and corrects errors through repeated cycles. The figure comes from a sponsored article and has not been independently confirmed for ChatGPT’s new allowances, but it illustrates why a few complex jobs can matter more than many short chats.
Pros and Cons of the combined system
Pros:
- Less navigation – Ideas, files, and completed actions sit closer together, so you need to switch between Chat and Work less often.
- Continuous workflows – A conversation can lead directly to a website, presentation, or supervised web task.
- Visible consumption – The planned allowance balance and reset time in Settings should make intensive work easier to schedule.
- Better use of simple models – Instant chats that do not count toward the allowance may preserve capacity for demanding work.
Cons:
- One shared bottleneck – Intensive Codex use can reduce the resources available for complex ChatGPT tasks, and the reverse is also true.
- Uncertain costs – Without published allowance sizes and extra-credit prices, you cannot fully plan what heavy usage will cost.
- Hidden complexity – A simpler interface may conceal how differently various models and work stages consume tokens.
- Possible waiting periods – Once you reach the limit, you may have to wait for a reset or accept additional spending.
The new interface removes one kind of complexity while making another more noticeable. The Decoder reports that, depending on the plan, users face two operating modes and a total of 11 model settings whose differences can be difficult even for experienced users to understand. A shared usage display helps, but it does not replace a clear explanation of which tasks consume which allowance.
Device limits make multiple subscriptions harder
In addition to the token allowance, OpenAI is introducing card and device restrictions for multiple paid accounts, according to a report on the new account limits. Users who connect too many subscriptions to the same payment card or device may no longer be able to buy another subscription. If a device reaches the permitted number of paid accounts, a “Device limit reached” message appears and the affected account can no longer be used on that device.
The affected plans reportedly include ChatGPT Go, Plus, Pro 100, Pro 200, and Pro 500. Free accounts and Business, Enterprise, and Edu plans do not count. The source does not give the exact number of paid accounts permitted per device, and signing out of one account reportedly does not automatically free a place for another.
This will particularly affect people who have switched between several cheaper paid accounts on one laptop after reaching a usage limit. OpenAI instead points them toward buying additional credits or waiting for the next reset. For someone with only one paid account, the shared token allowance is likely to be more relevant than the device restriction.
The policy also creates a clearer distinction for teams: several personal subscriptions used on one device are not the same as a Business or Edu plan intended for an organization. The supplied sources do not explain how the restriction applies to shared family computers, school devices, or rotating workstations. You should therefore not assume that simply signing out will immediately remove a block from a device.
New Codex integrations increase value and consumption
Codex is not becoming irrelevant just because the main ChatGPT interface is being consolidated. OpenAI is simultaneously highlighting new integrations, including a Codex plugin from Qodo. According to the provider’s brief description, Qodo brings company knowledge and coding standards into Codex to catch especially critical issues across repositories before a pull request is opened. A pull request is a proposal to review a code change and merge it into a project.
This is a concrete example of why Codex is more than an extra chat window. A team could check whether a change conflicts with internal standards or affects several projects before beginning the formal review. However, the claim that it catches the highest-priority issues comes from the provider and cannot be independently verified from the short summary.
Integrations of this kind may increase the value of Codex, but they can also trigger longer chains of work. The sponsored Technology Review article therefore argues that organizations should not count only tokens but compare their costs with changes that pass review and are actually merged. The underlying idea also applies to nontechnical users: what matters is not only how much allowance a task uses, but whether it produces a usable presentation, a working website, or a genuinely reviewed piece of work.
As cautionary examples, the article cites a report that Uber exhausted its entire 2026 Artificial Intelligence (AI) budget by April after expanding its use of Claude Code, as well as GitHub Copilot’s move from flat-rate pricing to metered credits. It also mentions a Gartner prediction that AI coding token costs will exceed average developer salaries by 2028. These are not claims about your ChatGPT subscription, but they indicate that intensive agentic use has different economics from occasional chatting.
What this means in practice
For beginners: Start with a clearly bounded assignment and then inspect the usage display in Settings. If you ask ChatGPT to create a presentation from several files, check the remaining allowance and reset time before starting a second substantial job. A non-counted Instant model may be sufficient for short questions if it provides the quality you need.
For advanced users: Deliberately separate basic questions, demanding document work, and Codex assignments. Compare not just the tokens consumed but the result: Was the presentation usable, the website completed, or the code change properly reviewed? Repeated correction cycles and misdirected agents can consume substantial capacity without producing a matching benefit.
If you use ChatGPT and Codex side by side for work, an internal order for large assignments may also help. Running a critical Codex job shortly before a reset could make more sense than using several experimental runs at the start of the allowance period. Multiple personal subscriptions on the same device are no longer a reliable workaround because card and device limits may block that route.
The sources provide no separate information about availability, prices, language support, or data-protection rules for Switzerland. They also do not explain how company files, internal knowledge, and source code are handled within the new integrations. Swiss individuals, schools, and companies can therefore draw conclusions about the general usage restrictions, but not make a country-specific privacy assessment from these reports.
The merger makes ChatGPT easier to understand at the interface level, but it ties general and specialized work more closely to the same allowance. Occasional users may barely notice, while people working with long documents, websites, or Codex workflows will have to treat the usage display as part of their planning. The unresolved risk is that the actual allowance and device thresholds remain unclear, along with whether simpler operation will be matched by transparent costs.
Sources
- OpenAI is finally merging ChatGPT’s Chat and Work tabs – Medium unknown, 2026-11-02
- ChatGPT legt Chat und Work zusammen und Nutzer müssen mehr mit ihren Token haushalten – Medium unknown, 2026-11-01
- ChatGPT setzt neue Gerätelimits: OpenAI macht Schluss mit Abo-Hopping – Medium unknown, 2026-11-01
- Stop counting tokens, start counting what ships – Medium unknown, 2026-11-02
- Qodo builds context into AI code review with Codex – Medium unknown, 2026-11-02
Image: Mikael Blomkvist via Pexels