ChatGPT and Codex shared usage limits from December: chat joins the allowance Codex already runs on

OpenAI has added a short notice to its Help Center, Understanding ChatGPT and Codex usage limits. Starting in December, Plus, Pro and Business plans get one usage allowance across ChatGPT and Codex, measured in tokens. It names no day and no allowance size. It does give a reason: OpenAI is bringing “the agentic capabilities of Work into the main ChatGPT experience.”

The quick reading is that Codex is losing its own meter. The rest of OpenAI’s documentation points the other way.

Is Codex or ChatGPT chat moving into the shared allowance?

Chat. Codex already shares. OpenAI’s guide to managing usage in Work and Codex says “Work and Codex share the usage allowance included with your plan,” and the Codex pricing page says Work in ChatGPT “uses the same pricing, credits, and usage limits as Codex.” The same guide describes chat as the part outside:

Chat has its own model availability and message limits. Work and Codex
share a separate usage allowance.

From December, chat draws on that allowance and is counted in tokens, like a Codex run already is. Codex is not untouched, though. The pool it shared with Work gets a new and much busier user.

What does OpenAI’s ChatGPT and Codex limits notice actually commit to?

Usage will depend on “the model you choose and the task you ask it to do.” Instant chat “will not draw from your usage.” Combined remaining usage and reset times will show in Settings → Usage, and at the limit you can wait, upgrade or buy credits “to continue using eligible features.” Then a reassurance:

We expect most users won't need to wait for their usage to reset more
often than they do now.

That is a forecast about the population. From your own account, with no allowance size published, you can’t test it.

What does the ChatGPT and Codex limits notice leave out?

Nearly every number. It doesn’t say how large the shared allowance is on any plan, or what happens to the windows. Today Plus and Business Standard have a documented five-hour limit in Work and Codex, weekly limits “may also apply,” and Pro has no five-hour limit. Enterprise and Edu go unmentioned. It also doesn’t say whether chat’s per-model message caps, such as the weekly GPT-6 Pro limit, go away or stay on top of the token pool. The same goes for the 2.5x rate of Fast mode and the 8x of Ultrafast: documented for Work and Codex, nothing said about chat. The other help pages, read the same day, still describe today’s arrangement and don’t mention December.

Who will notice the shared ChatGPT and Codex allowance most?

People who use heavy, non-Instant chat models and Codex on the same plan. Today those draw on different budgets. From December a long conversation on a heavy model and a Codex refactor draw on one. If the five-hour window survives the merge, which the notice doesn’t say, they share that too.

A few smaller effects follow from today’s rules. When a chat model hits its limit today, the Pro tiers page says to “use another available model.” For the pool, the Work and Codex guide says “Switching models does not restore allowance in a shared usage pool.” Instant chat is the one mode the notice exempts. A Codex review loop on your ChatGPT plan, which the Codex plugin for Claude Code warns “may drain usage limits quickly” (see how people split work between Claude Code and Codex), would now drain allowance your chat needs. And apps with plan usage switched on, like Notion through ChatGPT token sharing, already count eligible requests against Work and Codex usage, so they will compete with chat too.

What should a Codex user check before ChatGPT limits merge?

Record a rough baseline. After a typical working day, open Settings → Usage, or run /status in a Codex CLI session, and note the percentages, reset times, and which models and settings you used. It won’t isolate what chat costs once the allowance, windows or rates change, but it will show whether the same week of work still fits. Set a weekly cap on any app that uses your plan. And decide before you hit the limit whether credits are your overflow, because they cover eligible features only.

My read on ChatGPT and Codex’s shared token meter

Chat was never unmetered on these plans. It was metered in messages, a unit you could count and that hid what each answer cost. Tokens are closer to the cost, and you can’t count them yourself. Anthropic took a step in the same direction when it renamed Max 20x to Ultra and led with the weekly limit. The pools and sublimits differ, but neither lab publishes the size of what you get. It is the trade I described with Notion’s pricing: a price you can’t compute from published figures is an estimate.

In May I wrote that I run Claude for most work and Codex on a few tasks where it’s measurably better. For a split like that, one pool is simpler to watch than a pool and a separate message cap, if the cap really goes. What it costs is predictability, and a baseline from November is the closest thing to a check you’ll have.