OpenAI puts ChatGPT and Codex on one token meter from December

OpenAI will pool the usage that ChatGPT Plus, Pro and Business subscribers get across ChatGPT and its Codex coding agent from December, and measure it in tokens. Help center guidance updated on 1 November says the allowance “will be shared across ChatGPT and Codex as we bring the agentic capabilities of Work into the main ChatGPT experience”.

The guidance gives no allowance sizes for the merged pool and no date beyond the month. It says “more advanced models and complex tasks generally use more of your allowance”, and that Instant chat will not draw from it at all.

ChatGPT Business is priced per user per month and sold to teams of two or more, according to OpenAI’s pricing page. Its help pages describe the current Chat allowances as separate from usage in Work and Codex, the agent and coding tools where long tasks run.

From December one allowance has to cover both. That moves ChatGPT Business closer to a metered service: the per-user fee buys a token budget, the heaviest users of the most capable models spend it first, and paid credits cover what is left.

A Standard seat stretches by model

OpenAI’s own estimates show how far the same seat stretches depending on the model. On a Standard Business seat, its most capable model, GPT-6 Astra, allows an estimated 5–45 local messages in a five-hour window in Work and Codex, according to the company’s Business models and limits page. GPT-6 Sol allows 15–150, and the cheaper GPT-5.6 Luna allows 250–2,000. OpenAI says the ranges are estimates, not guaranteed counts, and that cloud chats can use more of an allowance than local messages.

Standard seats “include limited Astra usage within their existing Work and Codex allowance”, the page says. Premium seats get more included Work and Codex usage and no five-hour limit. In Chat, Business Standard includes 15 messages a month with GPT-6 Pro, which is powered by GPT-6 Astra, shared with GPT-5.6 Sol Pro. Business Premium includes 50 a week.

Model choice is only one variable. OpenAI’s guidance on Astra says larger inputs and outputs, higher reasoning settings, Fast mode and multi-step tasks can all raise usage. It also states: “Switching models does not restore allowance in a shared usage pool.”

Credits pick up the overflow

When a Business user exhausts the included usage, the feature is blocked unless the workspace holds credits, according to OpenAI’s guidance on flexible pricing. Workspace owners buy credits through billing settings, and the credits are pooled across all users in the workspace. They are valid for 12 months after purchase, and owners can set automatic recharge and usage alerts.

Credits are non-refundable unless the law, a contract or an approved exception requires otherwise, the same page says. For self-serve Business customers, unused credits may qualify for a refund after a verified billing error, confirmed unauthorized activity or a confirmed OpenAI service failure or incorrect account restriction.

A Business workspace pays in two ways. The seat fee is set per user per month. Credit spend rises with heavy use of Astra, Codex and long agent tasks, and from December those tasks will draw down the same pool as a seat’s everyday reasoning chats.

The seat price stops fixing the bill

OpenAI says it is “preparing carefully for this transition to help prevent disruption”. “We expect most users won’t need to wait for their usage to reset more often than they do now,” the guidance says. Combined remaining usage and reset times will appear on the Usage page in settings, and a user who hits the limit can wait for a reset, upgrade the plan or buy credits.

The users most likely to notice are those who already hand long coding or agent work to Codex and Work. Until December that work has drawn on its own allowance, separate from Chat. Once the pools merge, an afternoon of agent tasks on Astra can use up allowance that the same employee would otherwise spend on reasoning chats, and further Astra use moves onto credits if the workspace has them and its spending controls allow it.

OpenAI’s most capable model uses the most allowance per task, Instant chat uses none, and on a Standard seat the estimates run from 5–45 Astra messages to 250–2,000 Luna messages in the same five hours. Measuring the allowance in tokens passes more of the cost of the expensive models to the customers who use them most.

OpenAI has not yet published the size of the merged allowance for any plan. Until it does, a Business customer knows its seat fee for December but not how much Codex and Astra work that fee will cover, or how much will move onto credits.