The announcement came from the person who runs OpenAI’s Codex, the company’s coding agent, and it was pitched as good news delivered with an asterisk. The $200-a-month Pro subscription would reopen to new users tomorrow, he said. The way usage is counted would change with it.
What the change amounts to is a smaller allowance at the same price. Compared with the original $200 Pro plan, the new version corresponds to roughly half the API spending, the Codex product lead said — a cut that customers will notice if they push the service hard.
The executive framed the trade as a choice rather than a retreat. OpenAI did not want to compromise elsewhere, he said, and the company would not bring back the five-hour usage limit that had frustrated subscribers. Users could still draw on their full weekly allowance when they needed it.
The pivot is the latest turn in a subscription that has been redrawn repeatedly since it first appeared. OpenAI introduced ChatGPT Pro at $200 a month in December 2024, aimed at people doing serious work. It later added a $100 tier, and the $200 plan became the top of a two-step ladder.
Codex changed the calculus. As the coding agent became the center of the company’s developer story, Pro subscribers began using the plan less like a chatbot and more like a workstation. That heavier use is what the new usage math is designed to contain.
Codex itself is the reason the plan had to be rebalanced. The agent lets developers run long, autonomous coding sessions, and those sessions consume far more compute than a chat thread. A subscriber using Codex for an afternoon can burn through as much capacity as a casual user would in a month.
The economics behind the announcement point the same direction. The product lead said OpenAI expects subscribers to get more done over time as models become more efficient and API prices fall, so each dollar buys more value even if the raw allowance shrinks.
Analysts said the language reflects a company trying to manage a hard problem — how to sell access to a product whose costs keep dropping while keeping revenue per user from falling with them. A plan priced at $200 with half the included compute is, in effect, a price increase hidden inside a policy change.
OpenAI’s framing is that the improvement should outrun the cut. The product lead promised that users would get more work done over time, because efficiency gains and price reductions on the underlying models would flow through to subscribers. The value per dollar, he said, would keep rising.
He also teased additions that would not count against usage at all, saying more features would be added to the plan tomorrow without consuming allowance. What those features are, he declined to say — a familiar OpenAI move of withholding details until the launch itself.
The promise not to reinstate the five-hour limit matters because that limit had become a sore point. Subscribers who ran through their Codex allowance in a few heavy sessions complained publicly that a plan meant for serious work was throttling the people who used it most.
Under the new system, the weekly allowance is meant to be spent, not rationed against the clock. The company’s guarantee is that the quota can be used fully when a user needs it, a design that trades a hard time cap for a larger but finite pool of compute.
That design also gives OpenAI a cleaner way to meter demand. A time limit throttles everyone equally, including light users who rarely hit the ceiling; a usage allowance concentrates the cost on the heaviest users, who are also the ones most likely to pay for more.
The change lands in a market where the competition has shifted from who builds the strongest model to who can deliver it cheapest. Rival labs have been cutting prices on frontier models, and OpenAI’s own message — that efficiency gains get passed to subscribers — is aimed at that audience.
People familiar with the company’s thinking said the reworked plan is also a way to reopen the top tier without reopening the cost problem that closed it. When demand for Codex spiked, the heaviest users became the most expensive to serve, and the subscription had to be rebalanced.
The timing matters too. The change arrives the same week OpenAI’s annual developer event put the company’s agent products at the center of its story, and Codex sits at the middle of that push. How the Pro plan counts usage is now part of how the company prices its agent business.
The result is a subscription that costs the same and delivers less raw compute, with the promise that it will feel like more over time. Whether subscribers accept that trade will depend on whether the models really do get cheaper fast enough to make half of the old allowance feel like enough.
