One tab, one allowance: ChatGPT and Codex share a single usage pool from December

The bottom line: the interface is the small change, the billing is the big one

On November 2, 2026, OpenAI sent users an email containing two changes: one about interface — the Chat and Work tabs in the ChatGPT desktop app will merge into a single experience; the other about billing — starting in December, ChatGPT and Codex will draw from one shared usage allowance. TechCrunch reported the email the same day, and OpenAI's own Help Center page already states the December shared-allowance rule. Of the two changes, the first saves a mouse click. The second may change the day each month when your allowance runs out.

First, pin down the most common misunderstanding: Codex is not being merged. The email is explicit: "Codex will remain a separate experience in the ChatGPT desktop app." What merges is Chat and Work, both inside the ChatGPT side of the app. The coding tool stays its own section, with its workflows and history unchanged. Anyone reading "merge" as "three features become one" will misjudge the scope of this change.

Second, pin down the timeline. The merged experience has no firm general release date: a small group of users gets it first for feedback, then it rolls out widely. The shared allowance does have a firm month: from December, Plus, Pro, and Business plans will count ChatGPT and Codex usage against a single token allowance. Interface merge: "soon, for some." Allowance merge: "December." Do not blend them into one schedule.

Why today's desktop app confuses people

To understand the merge, look at the current structure. Per OpenAI's Help Center, the new ChatGPT desktop app actually has three parts: Chat for questions, search, and conversation; Work for research, analysis, and producing documents, spreadsheets, presentations, reports, and Sites; Codex for software development against local files, repositories, and terminals. Inside the ChatGPT side, you also flip a toggle between Chat and Work. TechCrunch's verdict is blunt: with that many sections, it is easy to forget where you are working and where exactly your project lives.

The structure is a scar left by feature growth. Chat, agentic work, and coding started as three separate product lines, each with its own entry point and metering. Crammed into one desktop app, they force users to answer a classification question — "is this a conversation or a task?" — before any work begins. The merge takes that question away from the user: you no longer decide in advance whether a request is a question or a job. You start talking, and continue through to having ChatGPT carry the work out, in one place — bringing files and plugins, asking it to build a website, create a presentation, or act on the web, while you review and guide its work.

Anthropic made a similar move in September, merging Claude's chat and Cowork into one screen. Both vendors are heading the same way: the line between answering and doing is leaving the navigation bar and moving into the model's own judgment. But there is a real difference: Anthropic merged chat with a general agent, while OpenAI deliberately left Codex out of the merge — because Codex touches local files, repository permissions, and sandboxes, boundaries that should not be diluted by a chat window's casual defaults.

The change worth doing arithmetic on: one allowance pool

The interface merge saves clicks. The allowance merge changes consumption math. Until now, ChatGPT conversation usage and Codex coding usage were metered separately. From December, both draw from one token allowance pool. OpenAI's Help Center states four rules: the change covers Plus, Pro, and Business plans; usage is measured in tokens, with more advanced models and more complex tasks consuming more; Instant chat does not draw from the allowance; and combined remaining usage and reset times will be visible in Settings, then Usage. If you hit the limit, the page lists three options: wait for the reset, upgrade the plan, or buy credits to keep using eligible features.

OpenAI's line in the email is that most users will not be affected, but "based on usage and model, some folks might need to wait for resets more frequently." Read that sentence in two halves. "Most users" means light users: short chats, an occasional Work document, an occasional Codex edit — one pool is plenty. For heavy users, the merge means trade-offs: a long Codex session can eat the day's chat budget, and an afternoon of dense Work tasks can leave the evening's coding short. Two pools used to be insulated from each other. Now they compete for the same water.

One piece of context makes the merge more sensitive: on October 29, OpenAI had just doubled the Work and Codex allowances for the Pro 100, Pro 200, and Pro 500 tiers and cancelled a planned cut to the Pro 200 tier (covered in this site's October 30 cycle). Users therefore enter the December shared pool with recently enlarged allowances. That is both good news and a warning: the pool got bigger, but it also became one pool. Judging December's impact requires adding your chat usage and your coding usage together, not just noting that "allowances doubled."

Several key numbers are still unpublished: the per-plan token allowance, the burn rate of each model, the exact switch date within December, how Free, Go, and Enterprise plans are treated, and whether web and mobile will follow the same merge. OpenAI has published none of these. Any claim that a particular plan will or will not be enough currently lacks a factual basis. The honest framing is to treat December as a month to be measured, not a month to be prejudged.

Three user types, one task each before December

Type one: chat-first, occasional coder. You are most likely among OpenAI's "most users." Your one pre-December task is to open Settings, then Usage, and look at your current split: how much is chat, how much is Codex. If the sum sits far from the cap, the merge is a pure interface simplification for you. Nothing else to do.

Type two: daily heavy Codex user. You are the most affected group. Before the December switch, audit a full week: record, even roughly, how much goes to ChatGPT conversations, Work tasks, and Codex sessions. Even without exact token counts, task counts and durations reveal the relative proportions. If Codex already dominates your usage, build in a buffer for December: break unimportant long tasks into smaller ones, keep exploratory questions in Instant chat (which does not draw from the allowance), and concentrate real Codex work in the window just after a reset. One more option is worth verifying for yourself in official documentation: contemporaneous reports say Codex can switch to usage-based API billing when an OpenAI API key is present in the environment, instead of drawing on the plan allowance. Do not migrate on the basis of a secondhand report alone — verify it first.

Type three: people who write internal guides for a team. Any documentation that says "use Chat for X and Work for Y" will need rewriting once the merged build reaches your accounts. More importantly, the billing language changes: where your team used to explain a "chat allowance" and a "Codex allowance" separately, December requires "one pool, total." Budget approvals, cost allocation, and usage-alert thresholds all need to be recalculated on the merged basis. Waiting until someone hits the cap mid-December to rewrite the docs is too late.

Three misreadings to avoid

Misreading one: "ChatGPT and Codex are merging, so coding moves into the chat window." No. The email says Codex stays a separate experience, and OpenAI's Help Center confirms its workflows and history remain separate from ChatGPT's. What merges is Chat and Work, not Chat and Codex.

Misreading two: "The merge doubles my allowance." No. A shared allowance is one pool drawn from two places, not two allowances added together. The October 29 doubling was a separate event; do not fold it into December's merge as "allowances doubled again."

Misreading three: "Everyone is forcibly switched in December." No. The interface merge goes to a small test group first with no published general date; the allowance sharing starts in December, but the exact switch day and per-plan coverage details remain unpublished. Reading "December" as "everyone switches on December 1" overstates what has been announced.

Verdict: navigation fixed, metering just starting to tighten

Seen against OpenAI's 2026 product line, the pattern is clear: interfaces keep getting simpler, metering keeps getting unified. Chat, Work, and Codex used to be three entrances and two ledgers; now the entrances are converging and the ledger is converging too. For OpenAI, unified metering is easier to explain and operate. For users, it removes a buffer that separate metering quietly provided — you can no longer count on leftover chat allowance to subsidize a coding overrun, because there is only one pool now.

That is not necessarily bad. One pool with one clear usage page is easier to manage than two pools with two fuzzy counters. The real question was never "merge or not" but "after the merge, how big is the pool, and how fast do different tasks drain it?" OpenAI has not published either number yet. Until it does, the best posture is neither panic nor indifference, but a personal usage audit: know how much your chat, Work, and Codex each consume, then let December's usage page show you the merged total. The day the interface merges, you may not even notice. The day your allowance runs out, you certainly will.

(Fact-check note: the merged experience and the email quotes follow TechCrunch's November 2, 2026 report; the December shared allowance, covered plans, token metering, Instant chat exclusion, Settings usage page, and the three options after hitting the limit follow OpenAI's Help Center page "Understanding ChatGPT and Codex usage limits" verbatim; per-plan token figures, per-model burn rates, and the exact December switch date have not been published by OpenAI and are marked here as unknown, not estimated; the October 29 allowance doubling is background from this site's October 30 cycle.)