For most paying subscribers, a long ChatGPT conversation has never cost anything you could see. That ends in December. OpenAI quietly posted a short notice on its help center over the weekend saying that ChatGPT usage limits on Plus, Pro and Business plans will be measured in tokens and drawn from one allowance that Codex, its coding agent, also uses.
The notice itself is five sentences long. It names no date beyond “December”, gives no allowance sizes and promises that “most users won’t need to wait for their usage to reset more often than they do now.” However, read alongside OpenAI’s existing Work and Codex documentation, it says a lot about who should start paying attention to their settings now.
Codex already ran on a meter. Chat is joining it
The change is easy to misread as Codex losing its own budget. In fact, Codex already shares. OpenAI’s guide to managing usage in Work and Codex states that “Work and Codex share the usage allowance included with your plan,” while Chat “has its own model availability and message limits.”
So the new part is Chat. OpenAI says the move comes “as we bring the agentic capabilities of Work into the main ChatGPT experience.” An email to subscribers, seen by BleepingComputer, adds that a combined Chat and Work experience will be tested with a small group of users in the coming weeks, and that Codex stays a separate experience in the ChatGPT desktop app.
In practice, that means a developer who runs Codex jobs all afternoon and then asks ChatGPT for a detailed research summary will be spending from the same pot. Until now, OpenAI counted those two habits separately.
What still sits outside ChatGPT usage limits
The most important line in the notice is the exemption: “Instant chat will not draw from your usage.” The subscriber email describes this as “Unlimited Instant” being included with the plan. As a result, quick everyday questions answered by the Instant model should not touch the shared pool at all.
OpenAI’s documentation lists a few other things that currently have their own meters. Regular voice conversations in Chat have their own limits and don’t consume Codex usage. Dots, OpenAI’s always-on assistants, also keep extended limits that sit apart from default usage “for now,” as the company said when it reversed a planned Pro cut on October 29. Codex sessions signed in with an API key also bypass the subscription and pay API rates.
The new notice doesn’t say whether those carve-outs survive December. It only spells out Instant.
The model picker now decides how long your week lasts
Because usage is counted in tokens, the model you pick will shape your ChatGPT usage limits far more than how many messages you send. OpenAI publishes rough ranges for Plus and Standard Business seats, estimating how many local Work or Codex messages fit into one five-hour window. The gap between the top and bottom of the range is striking.
| Model | Estimated messages per five-hour window (Plus) |
|---|---|
| GPT-6 Astra | 5–45 |
| GPT-6 Sol | 15–150 |
| GPT-6.1 Sol | 15–160 |
| GPT-6 Luna | 350–3,000 |
By our arithmetic, Luna stretches a window roughly 65 to 70 times further than Astra at both ends of those ranges. That is the real lever. For example, a Plus subscriber who leaves the most capable model selected for routine edits could hit a wall in a single afternoon, while the same work on a lighter model might never come close.
OpenAI also warns that switching models “does not restore allowance in a shared usage pool.” In other words, dropping to Luna after you run out won’t help; it only slows the drain from that point on.
A plan-by-plan read of the new ChatGPT usage limits
Plus and Business Standard. These seats have the tightest current structure: a five-hour window plus, in some cases, a weekly limit. Both must have room left for work to continue. Plus also includes only limited Astra usage, with credits available after that. These are the subscribers most likely to feel the new ChatGPT usage limits first.
Pro $100, $200 and $500. These tiers currently have no five-hour limit in Work and Codex, though an overall allowance still applies. On October 29, OpenAI doubled their Work and Codex limits after subscribers objected to a planned reduction for Pro $200. “We heard your kind feedback,” Tibo, who fronts ChatGPT and Codex on X, wrote at the time. Pro users therefore enter December with the most headroom.
Free, Go, Enterprise and Edu. The December notice doesn’t name any of these plans. Meanwhile, Enterprise already has token-based and credit-based billing options of its own, so the change appears aimed at self-serve subscribers first.
Fast mode, resets and other ways to burn the pool
Speed settings are the second hidden cost. OpenAI’s Codex speed documentation says Fast mode uses included subscription limits at 2.5 times the standard rate. Ultrafast, available only on Pro $500 and eligible Enterprise and Edu workspaces, runs at eight times the standard rate. If Chat inherits those multipliers under the shared ChatGPT usage limits, a habit of leaving Fast mode on would cost more than it does today.
Resets are the other piece to understand. Banked resets, which OpenAI handed out during the Astra launch in September, can expire. Purchased instant resets, offered to some Plus and Pro accounts, cannot be saved and move your weekly cycle forward. Once Chat shares the pool, any reset you use will cover both kinds of activity.
Questions the help page leaves open
- The start date. The notice says only “December.”
- The size of the pool. OpenAI does not publish numerical allowances for any tier, and it described its Pro doubling in relative terms.
- GPT-6 Pro in Chat. Its message limits are currently separate from the Work and Codex allowance. It is unclear whether those caps disappear or stack on top of the tokens.
- Speed multipliers in Chat. OpenAI documents Fast and Ultrafast rates for Work and Codex only.
Rivals are moving in a similar direction. Google, for instance, has been folding more agent-style tools into its paid tiers, as our look at Google’s September AI plan updates showed. Agents that act on your behalf, like the ones in Meta’s Muse launch, also cost far more to run than a chat reply. Token metering is a way to pass that cost through, and compute was the biggest bill in Anthropic’s IPO filing as well.
Four things to do before ChatGPT usage limits merge
- Record a baseline. Open Settings → Usage after a typical week of Work or Codex use and note how much is left. That gives you something to compare against once Chat joins.
- Default to Instant for quick questions. It is the only model OpenAI has promised will stay outside the pool.
- Match the model to the job. Save Astra and high reasoning effort for genuinely hard problems. OpenAI itself notes that Astra at low effort can beat Sol at high effort.
- Check Fast mode and saved resets. Turn off Fast mode for long jobs, and use any banked reset before it expires.
OpenAI’s promise that most people won’t notice may well hold for casual users, since Instant covers the bulk of everyday chatting. The subscribers to watch are heavy Plus users who code with Codex and research in ChatGPT on the same day. For them, the new ChatGPT usage limits turn two separate budgets into one, and the first test will be whether a normal week still fits.
