codex usage limits shared with chatgpt

Codex Usage Limits Are Moving Into the Same Pot as Your Chats. Here’s What’s Actually Confirmed.

TL;DR: OpenAI has told subscribers that “your use of ChatGPT and Codex will draw from the same allowance”. One pot, not two — so the agent writing your app and the chat window where you think out loud start competing for the same budget. Unlimited Instant is excluded. What is not confirmed is the date: the email carries none, and the widely repeated “December” comes from a help-centre notice we could not read. Your codex usage limits are changing; here are the three things worth doing before they do.

What OpenAI actually said about codex usage limits

On 1 November, BleepingComputer’s Mayank Parmar reported on an email OpenAI sent subscribers. The sentence that matters is short: “Your use of ChatGPT and Codex will draw from the same allowance.”

A subscriber posted the same email to Hacker News the night before, quoting two lines of it — “Chat and Work are coming together in one experience” and “A shared allowance for ChatGPT and Codex”. Two people, same wording, arriving separately. That is why we are reporting it.

Alongside the shared pot, here is what else the email says about codex usage limits:

  • Unlimited Instant does not count against the shared limit. It stays included in the subscription.
  • Heavier work costs more. More advanced models and more complex tasks consume more of the allowance.
  • There is one number to watch. Combined remaining usage and reset times appear under Settings → Usage.
  • When it runs out, you wait for the reset, upgrade, or buy credits for eligible features.
  • Codex stays its own thing in the ChatGPT desktop app — it is the meter being merged, not the product.
  • The combined Chat and Work experience is being tested “in the coming weeks” with a small group first.

Context from a few days earlier: on 29 October, Thibault Sottiaux announced that OpenAI was “doubling all usage limits for all Pro 100/200/500 tiers and Business Premium, effective today”, and that it “won’t be reducing usage limits for Pro 200”. So codex usage limits went up days before they were pooled.

What is not confirmed about the new codex usage limits

We are being deliberate about this, because a lot of what is circulating is not sourced.

The date. The email announcing the change to codex usage limits carries none. A great many write-ups say December, and trace it to a notice on OpenAI’s help centre. That page returns an error to us, so we have not read it, and we are not going to repeat a pricing date we cannot see. Treat December as likely but unverified.

Which plans. The reporting we can stand behind describes this as coming to Pro subscriptions. Summaries naming Plus, Pro and Business all trace back to the same unread notice — and at least one of those summaries is openly AI-generated.

The numbers. There are none. OpenAI does not publish exact numerical limits per tier, which is the whole problem with “we doubled it”: you cannot check.

Why this matters if you don’t code

Here is the plain version. Until now, your codex usage limits and your chat usage were two separate meters. Chat ran all day and mostly nobody thought about it; Codex burned through its own allowance when you set an agent loose on your app. One meter means the thinking and the building come out of the same budget.

For a non-developer that is a bigger change than it sounds, because of how people in this audience work. You do not sit down and write code for six hours. You go back and forth — ask what a thing means, paste an error, ask again, then send the agent off to make the change. Half of that is chat. Under one allowance, the half that felt free now costs you build time.

The people most exposed are the ones who chose ChatGPT because of the split. One Hacker News commenter who uses both Claude and Codex said plainly that Codex limits are the ones they hit, and that near-unlimited chat was the single thing keeping them on ChatGPT. Another, who mostly chats and rarely touches Codex, said they would stay until the change lands and then decide.

There is a reasonable counter-view in the same thread: in token terms, ordinary chat is a rounding error next to an agent grinding through a codebase, so most people will never notice. Both can be true. It depends entirely on which half of your usage is the heavy one — which is exactly what Settings → Usage will now tell you. It is also the only place your codex usage limits are visible at all.

If the phrase “token allowance” is doing a lot of work in that paragraph, our vibe coding glossary unpacks it.

Three things to do about your codex usage limits

  1. Find Settings → Usage and look at it now. Get a feel for your normal week before your codex usage limits are pooled, so you have something to compare against. This is the only number OpenAI gives you.
  2. Work out which half is heavy. Spend a few days noticing whether you burn allowance on long agent runs or on constant back-and-forth. If it is the agent, this change will barely register. If it is the chatting, your build budget is about to get smaller.
  3. Move the cheap work off the expensive meter. Rephrasing, brainstorming and “what does this error mean” do not need your most advanced model — and under one allowance they now compete with the work that does. Our piece on how a coding agent’s harness can triple the bill covers the same trap from the other direction, and hard spending caps are worth setting while you are in there.

Who should care (and who shouldn’t)

  • Building a revenue app: care. Agent runs are your main cost, and your codex usage limits now share a pot with everything else. Watch the usage screen for a fortnight before you commit to a deadline.
  • Internal tool for your business: care if the tool is on a Pro seat. Check with whoever pays the bill which tier you are on — the confirmed reporting covers Pro.
  • Website or landing page: barely. Occasional builds will not trouble the allowance either way.
  • Learning to build: this one is for you, oddly. Learners chat far more than they build, and that is the usage pattern most likely to feel the squeeze.
  • No idea yet: the 60-second Tool Finder quiz will point you at tools whose pricing you can actually predict.

Our take

Pooling codex usage limits in with chat is defensible — two allowances that reset on different clocks is a genuinely confusing way to sell a subscription, and one number in one place is better. What bothers us is the asymmetry: OpenAI publishes no numerical limits, so “we doubled it” and “now you share it” land on the customer as claims they have no way to verify. We would not switch over this — but we would screenshot your usage screen this week, because when the change lands you will want evidence of what normal looked like. If a vendor’s pricing can only be checked by watching your own meter, that is worth knowing before you build a business on it.

Not sure which AI build tool suits how you actually work? Take the 60-second Tool Finder quiz →

Some links on this page are affiliate links. If you buy through them we may earn a commission at no extra cost to you.

FAQ

Are ChatGPT and Codex usage limits really merging?

OpenAI told subscribers that “your use of ChatGPT and Codex will draw from the same allowance”, in an email reported by BleepingComputer on 1 November and quoted independently by a subscriber on Hacker News. The wording is OpenAI’s own. The effective date was not in that email.

When do the new codex usage limits start?

Not confirmed. Many write-ups say December, citing a notice on OpenAI’s help centre that we could not access, so we are not repeating the date as fact. The separate combined Chat and Work experience is being tested “in the coming weeks” with a small group of users.

Does ordinary chatting now eat my Codex allowance?

Under one shared allowance, yes — your codex usage limits and your chat usage come out of one pot, with an exception. Unlimited Instant is included in the subscription and does not count towards the shared limit. More advanced models and more complex tasks consume more of it, so a long agent run costs far more than a quick question.

Similar Posts