As of November 2026, OpenAI says that from December 2026 the usage allowance on ChatGPT Plus, Pro and Business is shared across ChatGPT and Codex. Coding with Codex and chatting in ChatGPT will draw on the same allowance, so heavy use of one can leave less for the other.
This guide covers what OpenAI has said, what it has not said, and a few options. For the details of your own plan, rely on OpenAI's help article on ChatGPT and Codex usage limits.
What is changing?
Three points, as of November 2026, from that help article:
- From December 2026, the usage allowance on ChatGPT Plus, Pro and Business is shared across ChatGPT and Codex.
- Usage is measured in tokens, so advanced models and agent tasks use more of the allowance than simpler work.
- OpenAI has not published how large the shared allowance is on each plan, or an exact start day.
The last point matters. Until the figures are published, nobody outside OpenAI can say how many Codex tasks or chat messages a plan covers, so be careful with any number you see quoted elsewhere.
How much of my allowance will Codex use?
It depends on your plan and the models you pick, and there is no reliable rule of thumb yet. The only thing OpenAI has said is that the pool is shared and that heavier models and agent tasks use more of it.
To get a feel for your own habits, note roughly how much a normal week of Codex work takes and how much a normal week of chat takes. The shared allowance now has to cover both. OpenAI's help article is the place to check what usage information your account shows.
What can I do if I hit the limit?
There are three options. None is right for everyone.
- Adjust how you use it. If you only occasionally come close, use lighter models for routine questions and keep heavier ones for work that needs them.
- Check OpenAI's options. The help article is where to look for what OpenAI offers if you run out. This guide does not quote plans or prices.
- Move everyday chat elsewhere. If Codex is what you want to protect, keep your ChatGPT allowance for coding and do general chat on a pay-as-you-go service.
NanthAI Edge is one example of the third option. Edge sends cloud chat through your own OpenRouter account. You pay OpenRouter's listed price per token and NanthAI adds no markup. OpenRouter and the model providers can still apply their own limits, and your OpenRouter balance has to cover what you use.
Pay-as-you-go is not automatically cheaper. For light or occasional chat it can cost less than a monthly plan, and for heavy daily use it can cost more. Edge shows your OpenRouter balance in the chat header. Turn on Show Advanced Stats in Settings › OpenRouter and it also shows the cost of each reply, so you can see the figures for your own use. Our guide to pay-as-you-go AI versus a subscription shows how to work that out.
Does running Codex through NanthAI Edge avoid the limit?
No. Local runtime, which is included with NanthAI Pro (a one-time £4.99 purchase), connects Codex on your Mac to Edge so you can send tasks and follow the work from your phone or browser. Codex still runs under your existing ChatGPT or Codex subscription, so its use still counts against your allowance.
Edge's site says provider charges and limits apply separately from the NanthAI purchase. For Codex, that means the shared allowance is OpenAI's, not ours. To see how Codex tasks from your phone work, read how to run Codex on your Mac from your phone.
Local runtime also supports Pi. Pi uses its own configured providers, subscriptions or local models, including OpenRouter and Ollama. If you set Pi up with OpenRouter or a model on your Mac, that work goes through those providers and not through your ChatGPT allowance, though their own charges and limits apply. Pi is a different tool from Codex, so this changes your workflow. It does not let you run Codex without using your allowance.
Try it
If you want everyday chat outside your ChatGPT allowance, open chat.nanthai.tech, connect your OpenRouter account and turn on Show Advanced Stats to watch what you spend. The Bring Your Own Key page covers the connection, the price transparency page explains the cost display, and the local runtime page covers Codex and Pi on your Mac.

