ChatGPT and Codex usage limits are being merged. Starting in December 2026, Plus, Pro and Business subscribers will draw from a single allowance for both the chatbot and OpenAI’s coding agent, measured in tokens. OpenAI has updated its help documentation accordingly. Regular Instant chats will not count. For developers who run Codex heavily and also chat with demanding models, the math is about to change.
The move is part of a bigger restructuring. OpenAI wants to bring the agentic capabilities of ChatGPT Work into the main ChatGPT experience, so that users can go from a simple conversation to a multi-step task without switching modes. The shared allowance is the billing side of that plan.
What changes for ChatGPT and Codex usage limits in December
The key statement is on OpenAI’s help page on usage limits: starting in December, the usage allowance of the Plus, Pro and Business plans will be shared across ChatGPT and Codex. Usage will be measured in tokens, the units of information a model processes and generates. How much of the allowance a request consumes depends on the model chosen and on the task. More advanced models and complex tasks generally use more, OpenAI says.
Instant chat is the exception. According to the help page, it will not draw from the allowance. People who mostly use ChatGPT for quick questions with the default model are therefore unlikely to notice much.
Combined remaining usage and reset times will be visible under Settings, Usage. Once the limit is reached, there are three options: wait for the reset, upgrade the plan, or buy credits to continue using eligible features. OpenAI says it is preparing the transition carefully and expects that most users will not need to wait for their usage to reset more often than they do now.
What OpenAI told subscribers by email
OpenAI has also notified subscribers by email, as reported by BleepingComputer, which has seen the message. According to the report, OpenAI will test a combined experience for the Chat and Work tabs in the coming weeks, initially with a small group of users. Users are meant to bring in files and plugins and ask ChatGPT to build a website, create a presentation or do things on the web, while reviewing and guiding its work.
Codex will remain a separate experience in the ChatGPT desktop app, the email says. So it is the allowance that is being merged, not the interface. BleepingComputer also reports that OpenAI already doubled the Work and Codex allowances for its Pro 100, Pro 200 and Pro 500 plans effective October 29.
How Codex limits work today
The idea of a shared pool is not entirely new. OpenAI’s help page on using Codex with a ChatGPT plan already states that on plans with shared allowances or credits, Codex, ChatGPT Work, ChatGPT for Excel and Workspace Agents draw from the same allowance and credit pool, where those features are available. What is new is that more demanding usage in regular ChatGPT will join that pool in December.
According to OpenAI, consumption depends on several factors: the model, where the task runs, task complexity, context, reasoning, speed and the tools used. A long-running task can use substantially more than a short request. Some things stay separate. File uploads, image generation and voice in ChatGPT have their own limits that do not apply to Codex. Codex’s automated safety checks (auto-review) are free when signed in with a ChatGPT account and do not count toward the limit.
Enterprise customers on token-based agreements are already billed in US dollars for Codex instead of credits. The new help page explicitly mentions only Plus, Pro and Business. It does not say what happens to Free, Go, Enterprise and Edu.
What this means for developers
For many developers, Codex has become a work tool that runs for hours. Until now, they could think about their coding budget separately from their chat budget. Going forward, the two compete. If you start long agent tasks in Codex during the day and discuss an architecture question with a reasoning model in the evening, both come out of the same account. How that feels in practice depends on the actual amounts, and OpenAI has not published them.
It fits the general direction. With models like GPT-6.1 Terra, OpenAI is betting on agentic use that consumes far more tokens than classic chat. Google also bills agent features such as Gemini Computer Use strictly by tokens. A single token allowance is the logical consequence for OpenAI: expensive tasks cost more of the allowance, no matter which interface they run in.
In practice, that means:
- Watch your usage: In the Codex CLI, the
/statuscommand shows the current state; in ChatGPT, it is Settings, Usage. If you use both tools from December, pay closer attention in the first few weeks. - Choose models deliberately: Instant chat, which does not draw from the allowance, is enough for simple questions. Reasoning models are worth it for tasks that actually need them.
- Plan long agent runs: Large refactorings or test runs in Codex consume a lot. If you know such a run is coming, avoid maxing out the allowance in chat at the same time.
- Know the alternatives: Codex can also be used with your own API key. In that case, usage is billed through the API rather than the ChatGPT subscription.
- Teams: In Business workspaces, admins should check which options for additional credits and role permissions apply before employees start hitting limits in December.
What is still open
Many details are missing. OpenAI gives no exact date in December and no concrete token amounts per plan. It is also unclear whether Codex’s existing five-hour and weekly windows will remain in their current form, which features exactly count as Instant chat, and which already consume allowance. OpenAI does not describe how chat tokens and Codex tokens are weighted against each other.
The way the change is being communicated also stands out. It appears on a help page and in an email; there is no official blog announcement so far. For a change that directly affects paying customers, that is rather thin.
Our take
The merger makes sense from OpenAI’s point of view. When Chat, Work and Codex use the same models and handle similarly demanding tasks, one token pool is easier to manage than separate limits. It can benefit users too, for example someone who barely uses Codex and can spend the allowance in chat instead. Heavy users of both, however, lose the certainty that a busy chat day will not eat into their coding budget. Whether OpenAI’s expectation that hardly anyone will hit resets more often holds up will only become clear in December. Until then, it pays to know your own usage. For how OpenAI wants its models to behave more broadly, see the latest Model Spec update.
