ChatGPT Usage Limit in December: One Allowance for Chat and Codex. What Teams Should Do Now
Starting in December, the ChatGPT Plus, Pro and Business plans share one allowance across ChatGPT and Codex, measured in tokens. Quick chats with Instant models don't count. OpenAI hasn't said how big the allowance is, so settle who on your team uses which model for what before the switch.
CellCog found no announcement on OpenAI's blog or on X. The change sits in a help article on usage limits that was updated on October 31. The Decoder reported on it on November 1. We could not open the help article directly, so our account rests on the verbatim quotes at CellCog and the report from The Decoder. If you use ChatGPT at work every day, this affects you directly, because it applies to every chat with a strong model.
What exactly changes in December?
Separate limits become one shared pool. Codex is OpenAI's coding agent, and Work is the mode in which ChatGPT carries out tasks on its own. Those two already share usage today. What's new is that from December, regular chats with the stronger models draw from the same pool.
- Who is affected: Plus, Pro and Business. The help article doesn't name Free, Go or Enterprise for this change.
- Unit: tokens, the chunks of text AI models calculate in. Demanding models and tasks use more.
- Exempt: simple quick chats with Instant models.
- Where you see it: remaining budget and reset times appear under Settings, Usage.
- When you hit the limit: wait, change plans or buy credits.
- Interface: Chat and Work are merging.
According to the quotes from the help article, OpenAI expects most users won't have to wait for a reset more often than they do today. There's no way to check that yet.
What is still unknown?
The numbers that matter most are missing. Still open:
- the exact start date in December
- the size of the allowance per plan
- how chat tokens and Codex tokens are weighed against each other
- what applies to Business seats with credit billing
- whether credits also cover regular chats with strong models
- whether the five-hour limits stay, which the pricing page still lists for Plus and Standard Business
- whether the limits raised in late October for Pro and Business Premium carry over into the new pool
OpenAI called that increase a doubling, but according to heise, a German tech news site, Pro 200 probably just stayed at its previous limit.
Why does model choice become a cost question?
Because the models drain the pool at very different speeds. There is no conversion for the new allowance, but today's credit rates for Codex and Work give you a good reference point. Credits are OpenAI's billing unit for usage-based costs. The model names here apply only to Codex and Work: the pricing page says GPT-6.1 Terra and Sol aren't available in Chat. We calculated the factor ourselves from the output rate.
| Model | Credits per 1M input tokens | Credits per 1M output tokens | Factor vs. Luna | Plus estimate per 5 hours (local messages) |
|---|---|---|---|---|
| GPT-6 Luna | 2.5 | 12.5 | 1 | 350 to 3,000 |
| GPT-6.1 Terra | 25 | 125 | 10 | 35 to 285 |
| GPT-6.1 Sol | 50 | 250 | 20 | 15 to 160 |
| GPT-6 Astra | 250 | 1,250 | 100 | 5 to 45 |
The estimates come from OpenAI, apply to Plus and Standard Business, and aren't fixed limits. Weekly limits may apply on top. Also keep in mind:
- Speed costs extra: Fast mode uses your subscription allowance 2.5 times as quickly as standard, Ultrafast 8 times.
- Images are expensive: image generation uses up your limit about 3 to 5 times faster on average, depending on quality and size.
- Older doesn't mean cheaper: GPT-5.6 Sol currently sits at 100 and 500 credits, double GPT-6.1 Sol. The pricing page also mentions a promotional price through at least November 21, without giving the amount.
What does this mean for you and your team?
You need a short model rule before December, or chance decides when you hit the limit. Here's how to go about it:
- Track usage. The usage dashboard only shows Work and Codex today, since regular chats don't count yet. So for two weeks, have people roughly note who chats with strong models and how often.
- Set a default model. OpenAI's pricing page doesn't map models to tasks. For heavy usage it only advises smaller models or narrower tasks. Our take: rewriting emails, summarizing and translating belong on Instant or the smallest model you have. Use the middle one for analysis and longer texts, and the strongest only for tasks where a mistake costs more than the usage.
- Separate coding from office work. Heavy Codex users can slow down their own chat from December on. In our view, a bigger plan will usually pay off for those people more than constant waiting.
- Cap credits. Decide who may buy more, and check in advance which purchase options your account or workspace offers.
- Set a date. As soon as OpenAI publishes the numbers, run the math and adjust the rule. Don't change plans because of the allowance until then.
This trend isn't new: Lovable and Figma also bill AI work through credits.
Are Make and n8n automations affected?
In our assessment no, as long as they run on an API key. The API is billed at its own API rates, and the quotes from the help article only name subscription plans as affected. The cost risk there is a different one: usage with no ceiling if you don't set one. Our comparison of Make vs. n8n shows how that adds up. If you want workflows that don't depend on a chat subscription, we're happy to help.
What about the new device limit?
It hits personal plans, not Business. The Decoder reported on November 1 that OpenAI now caps how many paid accounts can run on one device or with one card. This applies to Go, Plus and Pro, not to Free, Business, Enterprise and Edu. OpenAI gives no number, and logging out doesn't automatically free up a slot.
This matters for companies where people share a computer and each use their own Plus account: once the limit is reached, the account can no longer be used on that device. Business costs 20 dollars per user per month billed annually, 25 billed monthly, with a minimum of two users. On annual billing that matches Plus at 20 dollars, and you also get company sign-in (SAML SSO). Our take: if your company runs several personal Plus accounts, look into switching now. If you also want to settle how your team uses AI overall, take a look at our AI consulting for SMEs.
Frequently asked questions
Will ChatGPT get more expensive with this change?
The reports we read don't mention a price increase. If you work a lot with strong models, though, you may hit the limit sooner and then have to wait, change plans or buy credits.
Can I keep working when the allowance runs out?
For simple questions, yes, because quick chats with Instant models don't draw on the allowance. For everything else, OpenAI names three options: wait, change plans or buy credits. Whether credits also cover regular chats with strong models is still open.
Does the change also apply to ChatGPT Enterprise?
According to CellCog, the help article doesn't name Enterprise for this change. Per the pricing page, Enterprise with flexible pricing has no fixed limits anyway and bills through credits.
How many tokens does the new allowance include?
OpenAI hasn't published that so far, neither per plan nor as a conversion from today's limits. Any figures you read on this are estimates for now.
Sources
Tell us what you're planning. You'll get an honest assessment within 24 hours.
More articles