OpenAI is preparing a structural change to ChatGPT that goes well beyond a cosmetic interface refresh. The company will merge the Chat and Work tabs into a single experience, and starting in December 2026, usage of ChatGPT and Codex — its coding tool — will be deducted from the same allocation on the Plus, Pro and Business plans. In practice, time spent coding with Codex will eat into the margin available for chatting, and vice versa.
- Interface consolidation: ChatGPT’s separate Chat and Work tabs will be folded into one unified experience, while Codex remains a distinct experience inside the desktop app.
- Shared quota: From December 2026, ChatGPT and Codex usage will draw from a single pool on Plus, Pro and Business plans, ending the workaround of switching to Chat to preserve coding capacity.
- Cost control: The move gives OpenAI tighter control over compute consumption as conversational and agentic coding workloads converge.
What exactly is being unified
The two ChatGPT tabs, Chat and Work, will be merged into a single experience. The desktop application previously separated ChatGPT and Codex into distinct sections, and within ChatGPT it offered two tabs — a layout that users found confusing. OpenAI’s goal is to let users move from exploring an idea to having ChatGPT execute it without changing location: uploading files, asking it to build a website or a presentation, or running tasks across the internet while reviewing the work along the way.
Codex, however, will remain a separate experience within the desktop app. OpenAI will first test the new interface with a small group of users to gather feedback before rolling it out more broadly.
Why the shared limits matter
Until now, chatting and working did not draw from the same pool — and from December, they will. Work and Codex already shared an allocation, while chat models had their own limits. Under the change, operations involving models included in the new allocation will be deducted from it even when the user stays in Chat. More complex tasks and more advanced models consume a larger share.
Users face a twofold concern: not knowing how much headroom remains before hitting the limit — a critical issue when the tool is used for complete jobs rather than casual conversation — and losing the trick of preserving quota by switching to Chat, which ceases to be a savings route. OpenAI has not detailed the consumption equivalence per task or the capacity of each plan. Usage can be checked in Settings, and once exhausted, the options will be to wait, change plans or purchase credits.
What this reveals about OpenAI’s business
Limits are the lever the company uses to balance its compute accounts. Unifying the allocation lets it better control how much each user consumes as chat and agentic programming blend together. It is not an isolated gesture: on October 27, OpenAI halved the price of its GPT-6.1 Terra model for computer use, and it had previously expanded quotas on certain plans. All of it points in the same direction — adjusting the flow of usage without touching the headline price.
The backdrop is spending. In May, OpenAI president Greg Brockman said, according to Reuters, that the company expected to spend $50 billion (roughly €44.69 billion) on compute in 2026. With that bill, every token counts, and the shared quota is a way of distributing that cost among subscribers. It mirrors the same movement seen when GitHub Copilot shifted to usage-based charging in what became known as the “tokenpocalypse.”
“The question it does not answer is the only one that matters to the person paying: how many real tasks fit in my plan before I am left high and dry?”
The interface unification is genuinely good news — the two tabs made little sense and the application was a mess. But the redistribution of limits is what users will actually feel, and not in their favor. Merging the chat and code pools means an intense day of programming can leave no margin for anything else, and the lack of transparency about how much each task consumes turns usage into an unknown. OpenAI presents it as convenience, and in form it is; at its core, it is cost containment. As conversational and agentic workloads converge under one meter, subscribers will need to budget their AI time as carefully as they budget their compute bills.