OpenAI has told Plus, Pro and Business subscribers that ChatGPT and Codex will share one usage allowance starting in December. The allowance is measured in tokens, covers the agent features of ChatGPT Work as they move into the main ChatGPT experience, and does not charge Instant chat. That is the whole confirmed change: no token counts, no dates beyond "December", no per-plan numbers.
Because limits are the thing paying users feel most, this is worth reading carefully. Below we separate what OpenAI actually wrote from what people will infer, and walk through the history that explains why it is happening.
TL;DR: what OpenAI says, and what it does not
| Question | Answer, per the Help Center article |
|---|---|
| When? | "Starting in December" with no exact date |
| Which plans? | Plus, Pro and Business |
| What is shared? | One allowance across ChatGPT and Codex |
| How is it measured? | Tokens; depends on model and task |
| Does Instant chat count? | No, Instant chat will not draw from usage |
| Will I wait more often? | OpenAI "expect[s] most users won't need to wait for their usage to reset more often than they do now" |
| Where do I see it? | Settings, then Usage: combined remaining usage and reset times |
| If I hit the limit? | Wait, upgrade, or buy credits for eligible features |
| Published token numbers? | None in the article |
What exactly changed in the OpenAI Help Center?
The article, "Understanding ChatGPT and Codex usage limits", carried an "Updated: 8 hours ago" stamp when we read it on November 1, 2026. Its core sentence says that "Plus, Pro, and Business plans' usage allowance will be shared across ChatGPT and Codex as we bring the agentic capabilities of Work into the main ChatGPT experience."
Two pieces of context sit in that sentence. First, the merge is tied to a product change: the agent features that lived in ChatGPT Work are being folded into ChatGPT itself. Second, the shared pool follows from that: once the same agent runs in both places, two separate meters make less sense.
The article also says OpenAI is "preparing carefully for this transition to help prevent disruption to your workflows." That is a soft promise, and it is the part to watch when numbers arrive.
Why tokens, and why Instant chat is exempt
Measuring in tokens means a quick question to a small model and a multi-hour agent run against a flagship model no longer cost the same "message". OpenAI says plainly that "more advanced models and complex tasks generally use more of your allowance."
Excluding Instant chat is the notable carve-out. It keeps casual use free of the meter, so only the heavy, agentic and reasoning-model use competes for the pool. For a Plus user who mostly chats, that probably changes little. For someone who runs Codex for hours, the pool now also has to absorb whatever they do in ChatGPT with advanced models.
How we got here: a short history of Codex limits
The Codex limit story has been unusually volatile, which is why a "most users will not notice" line deserves scrutiny.
- In July 2026 the rolling 5-hour limit was removed and weekly resets were reshuffled.
- Later that month OpenAI nudged quotas up and brought the 5-hour cap back in a different form alongside GPT-5.6 Sol.
- On August 25 Codex lead Tibo Sottiaux said the 5-hour window returns for Plus while Pro tiers stayed exempt for the coming months.
- On September 26 an outage prompted a promised reset of paid limits.
Each step moved the goalposts for people who chose a plan on the strength of how generous it felt that week. A merged token pool is a more structural change than any of those tweaks, because it changes the unit of account, not just the size of the bucket.
The Help Center excerpt we read does not mention whether the 5-hour window, the weekly cap, or both survive in the combined system. Treat that as an open question and look for it in the next update.
How this compares with Anthropic's approach
Anthropic made a related move on its side in October, renaming the top Max tier to Ultra and removing the 5-hour limit for verified accounts while keeping a weekly cap and adding API credits. The two companies are converging on the same shape: a subscription that is really a weekly compute allowance, with credits as the overflow valve. OpenAI's version differs in that the chat product and the coding agent now draw from one tank.
| Aspect | OpenAI (announced, December) | Anthropic (October change) |
|---|---|---|
| Unit | Tokens, shared by ChatGPT and Codex | Weekly cap, 5-hour limit removed on top tier for verified accounts |
| Overflow | Wait, upgrade or buy credits | API credits |
| Free carve-out | Instant chat | Not applicable |
| Numbers published | None yet | See the linked post |
What to do before December
- Note your current habits. Record roughly how often you hit a Codex reset today. The company's promise is relative to now, so your baseline is the only way to check it.
- Find the Usage screen. OpenAI says combined remaining usage and reset times will live in Settings, then Usage. Open it when the change lands and screenshot it.
- Choose the model deliberately. Because advanced models burn more of the pool, route routine work to lighter models and keep flagship models for tasks that need them. Our ChatGPT Work versus Codex guide explains where each surface fits.
- Decide your overflow plan now. If you depend on long Codex runs, price credits and the next plan up before you are mid-task and blocked.
- Watch for fine print. Enterprise, Edu and Free plans are not named in the passage we read.
Who gains, who loses: three user profiles
The honest answer is that it depends on your mix, and OpenAI has not published enough to compute it. Here is how the shape of the change plays out for three kinds of subscriber, as our reading rather than an OpenAI statement.
The chat-first Plus user. If you mostly use Instant chat, nothing should move, because that mode does not draw from the pool. If you lean on advanced reasoning models for writing and research, you now spend from the same tank as any coding agent you run, which is new.
The Codex-heavy builder. You are the person most exposed. Before, Codex had its own meter, and whatever you did in chat did not touch it. After, long agent runs and reasoning-heavy chats compete. If you had been using a cheaper model in Codex to stretch the limit, that habit becomes even more valuable.
The Business team. Teams buy seats expecting predictable access. A single allowance per seat is simpler to explain, but the lack of published numbers makes budgeting hard. Ask your admin whether credits can be pooled across seats and whether the Usage screen exposes per-user consumption.
A usage meter beside a tap, illustrating one shared ChatGPT and Codex usage allowance
What people will ask next
Does buying credits fix it? The article says credits let you continue using "eligible features", which implies some features may not be creditable. The wording is the thing to check once the full policy is live.
Is this a price increase? Nothing in the article changes plan prices. Whether the combined token budget is larger or smaller than today's separate limits is exactly the number that is missing.
Should I switch to a competitor? Do not decide on a rumor. Compare actual allowances once published, using the official plan pages, and weigh the habits of your own workflow, such as the long agent runs discussed in our Codex versus Claude Code limits coverage.
What is claimed versus what is verified
The change itself is verified: it is in OpenAI's own Help Center, on the official domain. The soft claims are not verifiable yet. "Most users won't need to wait more often" is an expectation with no data attached, and no token allowance per plan has been published. Until those arrive, any specific number you see on social media for the combined limits should be treated as speculation.
We found no matching OpenAI blog post or release note in this pass, only the Help Center article, and a Hacker News submission linking it drew little discussion at the time of writing. We will add an update section to this post when OpenAI publishes the allowance numbers.
Why this matters beyond one plan page
Subscriptions for AI coding agents have been sold as flat fees for an open-ended activity. Compute does not work that way, so every lab keeps adjusting. Moving to a single token meter makes the economics honest for heavy users and simple for light ones, but it also makes it easier to tighten quietly: a change in how many tokens a task consumes can shrink your effective allowance without the headline number moving. That is the thing to monitor, and the reason to keep your own usage notes.
For teams on Business plans, the shared pool also means one person's agent-heavy week can eat into allowance that used to be earmarked, in effect, for chat. Ask admins whether per-seat visibility will exist in the new Usage screen.
Related reading
- ChatGPT Work vs Codex: what actually changes
- Codex Plus gets the 5-hour limit back
- GPT-5.6 Sol usage limits
- Codex and ChatGPT Work outage and limit reset
- Claude Ultra plan and weekly limits
- OpenAI Help Center: Understanding ChatGPT and Codex usage limits
- ChatGPT plans and pricing and Claude plans, the official pages to compare current tiers
- Hacker News discussion
Details reflect OpenAI's Help Center as of November 1, 2026 and may change before the December rollout.