The Era of Shared Quotas Arrives at OpenAI: ChatGPT and Codex Merge, Unlimited Prompting Ends

OpenAI recently quietly updated a Help Center document titled "Understanding ChatGPT and Codex Usage Limits." This seemingly minor change may mark the end of the "unlimited prompting" era—starting in December, usage quotas for Plus, Pro, and Business plans will be shared between ChatGPT and Codex.

A Three-in-One Change: Chat, Agents, and a Token Pool

According to the help document, this adjustment involves three things at once: first, ChatGPT and OpenAI's coding agent Codex no longer have separate quotas and instead share a single token pool; second, the previously separate Work agent mode is folded into the regular ChatGPT experience; third, quotas are measured in tokens rather than message counts. The document states plainly: "More advanced models and complex tasks typically consume more quota." In other words, having a large model refactor an entire codebase costs far more than asking it "what rhymes with orange."

The only exception is Instant chat mode. The document explicitly states: "Instant chat does not consume quota." Lightweight, quick Q&A is unaffected, but reasoning models, agent tasks, and Codex sessions will all draw from the same wallet.

Token is the basic unit of text processed by large language models and the basis for API billing and usage limits. Models don't process input by "characters" or "words"; they first split text into tokens—in English, one token roughly corresponds to 0.75 words, while in Chinese, one character or word often corresponds to one to two tokens. The cost of a conversation is determined jointly by input tokens and output tokens; the more advanced the model, the longer the context, and the more reasoning steps, the more tokens are consumed. This is the technical meaning behind the document's statement that "more advanced models and complex tasks typically consume more quota": token-based billing is closer to actual compute costs than per-message billing, but it also makes it harder for users to intuitively estimate their remaining quota.

Codex is OpenAI's coding agent product, capable of reading codebases, writing and modifying code, running tests, and completing multi-step programming tasks. Unlike ordinary chat, coding agents often need to repeatedly read large numbers of files, perform long chains of reasoning, and generate large blocks of code—a single session can consume several times or even dozens of times the tokens of ordinary Q&A. Therefore, putting Codex and ChatGPT into the same token pool has a far greater practical impact on heavy coding users than on light chat users.

User Anxiety: The Chain Reaction of a Shared Budget

The news didn't come from a launch event or keynote, but from a help page edit. AI tracking account Token Gremlin posted about the change on Saturday, racking up over 260,000 views; The Decoder subsequently reported it under the headline "OpenAI's chat and work merger marks the end of unlimited prompt pricing."

On the OpenAI developer community forum, a post titled "Usage quotas will be shared between ChatGPT and Codex" quickly drew a flood of complaints. One Plus subscriber wrote: "No, unlimited Instant doesn't make up for this." Another user said: "I don't want to budget an important conversation together with my next book project." That user also demanded that OpenAI disclose the actual included capacity along with a before-and-after comparison.

The core concern is straightforward: previously, if Codex quota was exhausted during a long debugging session, ordinary ChatGPT conversations remained safe; after December, the same bug could eat into the chat budget too. As one user put it, it's like a roommate's gaming PC and your fridge sharing a single power outlet—some lasagna is bound to get warm.

Agent Mode refers to AI not just answering questions but autonomously planning and executing multi-step tasks, such as browsing the web, operating tools, and invoking code execution environments. Compared with single-turn Q&A, agent tasks require continuous environmental feedback and intermediate reasoning, and token consumption accumulates rapidly with task complexity. The original article's mention that the Work agent mode is folded into the regular ChatGPT experience means such high-consumption tasks no longer have a separate budget but compete for the same quota as everyday chat. This explains the technical root of user anxiety: coding and agent tasks are inherently highly variable, and a single deep debugging session can instantly drain the shared pool—while ordinary chat users lack both awareness of and control over this.

Product Logic and Industry Trends

The Decoder notes that ChatGPT users currently must choose between two operation modes and 11 model settings. The 11 settings feel more like a cockpit than a product; merging chat and work aims to simplify choices. Meanwhile, Anthropic has already integrated chat and Cowork in Claude, and the industry is trending toward "one assistant that can both converse and get work done, running on a single meter."

OpenAI reassures users on the help page: "We expect most users won't need to wait for usage resets more frequently than they do now." Users can view their combined remaining quota and reset time on the "Usage" page in settings. Upon hitting the limit, there are three options: wait for a reset, upgrade the plan, or purchase credits—jokingly called the "holy trinity" of software subscriptions.

Anthropic's Claude is one of OpenAI's main competitors, and its product line likewise includes general chat and work-oriented agent capabilities (such as Cowork). The industry's "convergence of chat and work agents" is not merely interface simplification but a shift in business model: the old "unlimited prompting" subscriptions implicitly subsidized light users, but with the spread of reasoning models and agent tasks, the compute cost structure has changed, and unified token-based quotas have become a more sustainable pricing approach. Anthropic's similar integration shows this isn't just OpenAI's strategy but a signal that the entire AI assistant industry is transitioning from "message subscription" to "usage metering."

Unknown Pool Size and Official Signals

The help page does not specify the exact size of the new pool—no token figures, no specific December date. Paying users are told that chat and coding agents will share a budget, but outsiders cannot yet calculate whether this budget is larger, smaller, or roughly the same as before.

On the other side, OpenAI's Tebow (whose bio reads "Codex and ChatGPT at OpenAI") posted a series of countdown posts on X, announcing multi-account switching on mobile and hinting that "there will be resets in the next two days. Banked. You decide." The three posts together garnered about 5 million views. To be clear, there is no evidence that these resets are directly related to OpenAI's quota changes.

When AI starts being billed by token like utilities, the user's choice becomes simple: use it more carefully, or switch to a competitor that still offers an all-you-can-eat buffet?

One-line Summary

OpenAI merges ChatGPT and Codex quotas into a shared token pool, potentially ending the unlimited prompting era.

Paragraph Summary

OpenAI quietly updated its help document, announcing that starting in December, usage for Plus, Pro, and Business plans will merge ChatGPT and Codex into a single token pool, while Work agent mode is folded into regular ChatGPT, with quotas measured in tokens rather than message counts. The only exception is Instant chat mode, which doesn't consume quota. After the news was exposed by an AI tracking account, it triggered widespread user complaints, with the core concern being that coding tasks draining the quota will affect everyday chat, and that the official announcement lacks specific capacity and effective dates, leaving users unable to assess the real impact. The article notes that this adjustment aligns with the industry trend of merging chat and work agents—Anthropic's Claude has already adopted similar integration—signaling that token-based billing is replacing unlimited-prompt subscriptions.