OpenAI merges ChatGPT and Codex quotas, billing by token starting in December

Over the past four weeks, OpenAI's Codex users received a quota reset almost every other day. In the final days of the giveaway, the ledger tightened.

OpenAI's Help Center has just published a document titled "Understanding ChatGPT and Codex usage limits," and the very first sentence explains the change: starting in December, the usage quotas for the three paid plans—Plus, Pro, and Business—will be shared between ChatGPT and Codex. OpenAI's stated reason is that it is merging ChatGPT Work's agent capabilities into the main ChatGPT experience.

In plain terms: previously, a ChatGPT subscription gave you separate quotas for chatting on the web and writing code in Codex, with no interference between them; after December, both will draw from the same pool. The bigger change is in how usage is measured—quotas will be calculated uniformly by tokens rather than by message count. The only thing unaffected is Instant chat.

Billed by token, with only Instant as the exception

According to the help document, users' usage in both ChatGPT and Codex will be measured in tokens. How much quota is consumed depends on the model the user selects and the task given to it—the more advanced the model and the more complex the task, the more is deducted. Instant chat is not counted toward the quota at all.

OpenAI also offers reassurance in the document, saying it is preparing carefully for this transition and expects most users will not be greatly affected—the frequency of waiting for quota resets will not be higher than it is now. After the merge, users can view their remaining total quota and reset time under "Settings → Usage."

When the quota runs out, there are three options: wait for a reset, upgrade the plan, or purchase credits to continue using eligible features.

However, the document does not give a specific launch date, nor does it disclose how many tokens each plan tier will actually get.

There were signs of this earlier. On the morning of the 30th, Tibor Blaho, who has long tracked OpenAI product developments, spotted a new notice on the ChatGPT web checkout page: "Starting in December, plan usage will share a single limit across all ChatGPT and Codex features." He judged this to be related to OpenAI's ongoing merger of Chat and Work.

The day before, Codex lead Tibo (Thibault Sottiaux), in a reply to a user, had said the team was working on this but had not yet started rolling it out.

It is worth noting that, according to OpenAI's earlier explanation, Work and Codex already shared the same quota. So this adjustment is not creating a new pool out of thin air, but rather folding ChatGPT's everyday chatting into it as well.

Multiple accounts on one device may get locked

Beyond the quota merge, another rule tightened at the same time.

According to a summary by the Codex Update account, OpenAI has newly published a help page about bank cards and device restrictions. There is a cap on the number of paid accounts a single device can use, covering the Go, Plus, and Pro 100/200/500 tiers. If the cap is exceeded, users will see a "Device limit reached" prompt at login, or be logged out during use, even if payment has already succeeded. Logging out of another account will not free up a slot. Business, Enterprise, Edu, and free accounts are not counted toward this limit.

The timing is rather delicate. In the same week this rule took effect, Codex had just launched multi-account switching on mobile. Many heavy users' previous approach was precisely to buy several subscriptions and rotate through them—now that path has been explicitly blocked. Codex Update also reminds users that the officially sanctioned way to expand capacity is to purchase credits in the Codex app under "Usage & Billing."

"I might buy more Claude"

After the news spread, the community's reaction was not optimistic.

For many Plus users, the most attractive part of a ChatGPT subscription is precisely that web chatting has a separate, nearly unlimited quota. 𝕏 user xikhar said outright that the reason he likes the ChatGPT subscription is the independent chat quota; now that it is being merged with Codex, unless OpenAI quickly delivers a better model, he might switch to buying more Claude.

Some also noticed a detail: just a few days earlier, OpenAI had doubled the quotas for Pro and Business Premium, but Plus was not included. Under Tibor Blaho's post, some netizens complained that Plus's only perk is about to disappear, yet its quota did not double along with Pro's.

As for which models count as Instant and are not metered, there is currently only speculation. Netizen Angaisb_ speculated that perhaps only models like GPT-6 Sol and GPT-6 Luna remain unlimited. There has been no official response to this claim.

Some users have also done the math: only Instant chat is unrestricted; once you hit the limit, you can only wait, upgrade, or buy credits, and heavy Codex users will feel the pressure first.

Tibo's 28 days, from "The Reset Company" to "The Token Company"

To understand the community's emotional swing, we first need to look back at what happened in the Codex community over the past four weeks.

On October 4, Tibo made a pledge on 𝕏: for the next 28 days, he would either ship one improvement clearly useful to most Codex/Work users every day, or do a full quota reset. The challenge ran from October 5 to November 1, and the community soon spawned a batch of websites dedicated to tracking his daily updates; some even opened polls letting users guess whether the day would be an "improvement day" or a "reset day."

Those 28 days delivered quite a lot. On day one, the default speed of GPT-6 Astra and GPT-6.1 Sol in subscription products increased by about 50%; on day 7, GPT-6.1 Sol's default context expanded from 272k to 500k; on days 12 and 16, caching and request routing were optimized consecutively, effectively letting the same quota do more work; on day 22, GPT-6.1 Terra returned, and Tibo said it was clearly better at writing, persona, and understanding the nuances of requests, while consuming only half the usage of 6.1 Sol; on day 24, users could message their Dot and leave voice messages, just like texting a friend.

Of course, there were also plenty of resets. According to community tracking sites, across the whole period there were more than ten global resets and banked resets (resets saved for later use) combined. Tibo's announcements also became increasingly casual: on day 10 he said "rumors of my reset button being lost have been greatly exaggerated"; on day 23 he first announced the launch of the highly requested iCloud plugin, then immediately followed with "just kidding, today is a reset."

Image generated by ChatGPT

Not everyone was buying it. Some paying users complained that a global reset sounds nice, but it wipes out users' remaining quota and restarts the 7-day timer, which for people carefully planning their weekly quota actually disrupts their rhythm.

The climax came on day 25. Tibo announced that all quotas for the Pro 100/200/500 tiers and Business Premium would double starting immediately, and said OpenAI had listened to user feedback and would no longer cut Pro 200's quota.

The background here is that OpenAI had previously notified users that starting October 30, Pro 200's quota in Work and Codex would drop from 20 times Plus to 10 times Plus. Machine Heart had also previously reported on the controversy sparked by this round of adjustments; see "OpenAI Will Reopen the $200 Plan, but Tibo Got Slammed." Subsequently, Tibo added: "Graduating from The Reset Company, entering The Token Company."

On day 26, multi-account switching launched on mobile, and Tibo previewed that the next two days would both be banked resets; on day 27, he posted "Banked 1/2."

Looking back now, that joke about "The Token Company" was almost a preview of the quota merger.

"Unlimited" was itself a kind of subsidy

Putting the two things together, OpenAI's thinking is not hard to understand.

Limiting by message count essentially hides the cost of each question from users. Asking about the weather and having an advanced model reason for half an hour might be treated the same in terms of quota, with OpenAI absorbing the cost difference itself. After switching to token-based billing, a subscription is more like a prepaid package of API quota: you're charged for what you use, and any excess is topped up with credits, transferring compute costs into users' perception with explicit price tags.

Behind this is sustained compute pressure. After GPT-6 Astra launched in September, demand surged, and OpenAI temporarily paused new subscriptions to the $200 Pro plan; when it reopened, Tibo admitted that the new plan, converted to API spending, was only half of the old version. Frequent global resets also somewhat indicate that quota itself is a valve OpenAI can adjust at any time. And this time, the way the valve is adjusted has changed from "give one more" to "unified pricing."

For light users, the impact may indeed be limited; Instant can still be used freely. But for heavy users who discuss problems with advanced models in the web interface during the day and run long tasks in Codex at night, chatting and coding will from now on compete for quota in the same pool. The so-called "unlimited questions" will in the future exist only in the Instant tier.

This article comes from the WeChat public account "Machine Heart" (ID: almosthuman2014), author: the one who pays attention to quotas, published by 36Kr with authorization.