OpenAI will pool ChatGPT and Codex token allowances from December
From December, OpenAI’s Plus, Pro and Business plans will share one token allowance across ChatGPT and Codex.
Tokens measure the information a model processes and generates, so more complex models and tasks consume the budget faster. Settings will show the remaining balance and reset times. At the limit, users can wait for a reset, upgrade their plan or buy extra credits. Instant-model chats sit outside the allowance. OpenAI
The pooling comes with OpenAI’s integration of Work’s agentic capabilities into the main ChatGPT interface. For someone building and selling a product alone, ChatGPT tasks and Codex development work will draw on the same allowance. More complex work in either leaves less for the other. A fixed subscription price no longer supports the assumption that you can keep assigning agent work without deciding where the budget goes.
The Decoder expects heavy users on cheaper plans to receive lower-quality results as they exhaust their budget before the month ends. It also reports that OpenAI recently doubled limits on higher-priced plans, reading that increase as preparation for the merger. Both the quality forecast and the explanation for the increase are the outlet’s interpretations. The Decoder
The reporting provides no per-plan token figure, so nobody can yet say how many agent runs Plus or Pro will carry in December, or which tier covers a solo builder’s month.
December’s token allowance makes the cost of finishing a task matter
With usage counted in tokens instead of messages, a cheap call says little about the cost of usable work. A cheaper model at low reasoning effort that needs three retries can consume more of the shared allowance than an expensive run that finishes the job. Those retries belong in the same comparison. The useful number is total tokens spent getting one task finished.
Instant changes that comparison because a task completed there leaves the shared allowance untouched. Whether Instant is sufficient for a job becomes a budget decision. If its result needs another model to finish it, that later usage still belongs to the task’s total. Starting with the lightest model does not automatically save allowance.
Until OpenAI publishes per-plan figures, counting tokens per completed task is the only basis for estimating your month’s needs. That count is what you compare against the Plus or Pro allowance once those figures arrive, and it already includes the retries. It also separates work that can wait for a reset from work worth buying credits for.
The rest of today’s news
- ChatGPT card and device limits take effect: OpenAI now caps how many subscriptions one card can carry and how many paid accounts one device can hold. Free, Business, Enterprise and Edu plans are exempt, capped accounts stop working on that device, and logging out does not free a slot. The Decoder