OpenAI Is Merging ChatGPT and Codex Into One Token Pool in December

OpenAI is doing away with the separate ChatGPT and Codex allowances. Starting in December, Plus, Pro, and Business subscribers will draw on a single shared pool, and consumption will be measured in tokens rather than messages. The move lands roughly a week after OpenAI reversed a planned Pro cut and doubled every paid tier — and it reframes what a subscription actually buys.

What the December Change Does

A new support document titled "Understanding ChatGPT and Codex Usage Limits" lays out four points.

• Merged pool. From December, ChatGPT and Codex share one quota across Plus, Pro, and Business plans.
• Token-based metering. Consumption depends on which model is called and how heavy the work is. Higher-tier models, deeper reasoning, and long code tasks draw more quota.
• Instant chat stays free. Everyday "instant" conversation remains unlimited and does not count against the pool.
• Three exits when the pool runs dry. Wait for the reset, upgrade to a higher tier, or buy additional credits.

OpenAI has not published the actual token numbers or the per-model consumption ratios for each tier. The company's stated reassurance is that, for most users, the frequency of hitting a wall will not rise above today's levels. Heavy users are already doing their own arithmetic: a day spent on deep document analysis and agent runs in ChatGPT can mean Codex is dry by the time they open the desktop client in the evening.

Why the Pool Is Merging Now

The change is the third move in a week and a half. On October 30, Codex lead Tibo reversed a planned Pro 200 cut and doubled every Pro tier plus Business Premium. That was followed by a preview of a unified Chat and Work interface: files and plugins will be pulled into the same surface, with Codex kept as a separate entry point inside the ChatGPT desktop app. OpenAI president Greg Brockman has already conceded that the current app feels messy, and the separate Work tab is going away by year-end.

The sequence is legible: double the pool in late October to calm the backlash, test the unified interface in November to build the habit of calling agents from one place, and put ChatGPT and Codex on one token meter in December.

Anthropic Already Runs on One Pool

The same model already exists at Anthropic. Claude Pro and Max share one allowance across the chat interface and Claude Code: a long web session in the morning means less terminal work in the evening. Anthropic also just re-tiers its own plans: Max 5x becomes Max at $100 a month, and Max 20x becomes Ultra at $200 a month. Verified Ultra accounts drop the 5-hour rolling cap, but the weekly total stays fixed.

The comparison is the point. Both companies are converging on a subscription that behaves like a metered compute wallet rather than a flat "unlimited" chat plan.

The Device Limit That Kills Multi-Account Stacking

In the same week, OpenAI quietly added a hard cap on the number of paid accounts a single device can log into. The cap covers Go, Plus, and the Pro 100 / 200 / 500 personal tiers. Business, Enterprise, Edu, and Free accounts do not count against it.

Once a device hits the cap, new logins are blocked with a "Device limit reached" message, and active sessions can be kicked mid-use — even if the subscription was paid for. Logging out of an existing account does not free a slot; the device registration persists. This closes the common workaround of running several $20 Plus or low-tier Pro accounts on one dev machine and switching when each quota runs out.

The Reset Company Graduates

Tibo's most recognizable move over the past few months has been the free quota reset whenever the community complained. That era is ending. The December model replaces manual resets with a measured ledger: every token has a price, every call is recorded, and the subscription buys a compute wallet rather than an open-ended allowance. For heavy developers, that means reaching for a higher-tier model now comes with an explicit arithmetic step before they hit send.