Build2 publishers3 min readPublished
OpenAI's Premium seat charges exactly five times Standard for five times the usage
The included capacity is priced flat across both tiers, so the thing a heavy user is really buying is the end of the five-hour session ceiling.
The Engineer · Build desk
What happened
- OpenAI said on X on August 25th that Premium seats are now available inside the self-serve ChatGPT Business plan.
- Premium is $125 per user per month on monthly billing and $100 on an annual plan, against $25 and $20 for a Standard seat.
- A single workspace can hold both seat types, must keep at least two paid seats, and is capped at 200 paid seats in total.
- An introductory offer of $100 in workspace credits per Premium seat, capped at $500, closed before the wider release.
Compiled by The EngineerSomething wrong?How this is made
Why it matters
- cost Ten heavy users promoted to Premium on annual billing costs $9,600 a year more than leaving them on Standard, and that is recurring, not a one-off.
- constraint With no countable unit for included usage, an admin cannot model the upgrade before buying it and has to defend it by naming individuals instead.
- decision The live question is which named people sit on the expensive side of one workspace, which makes this a management judgement rather than a procurement one.
- precedent Pricing five times the usage at five times the money sets the expectation that OpenAI's next tier arrives as another multiplier rather than as new capability.
The multiplier on the price and the multiplier on the usage are the same number. At annual rates one Premium seat costs what five Standard seats cost, and it carries five times one Standard seat's included allowance [17]. Included capacity is priced flat across the two tiers, and concentrating a workload on one person earns no discount. What the higher seat adds that money cannot otherwise buy is the end of the rolling five-hour window, which lets one person spend a weekly allowance in long sessions rather than in five-hour slices [5]. That is a scheduling property. It matters to somebody running sustained Codex jobs, which is the work OpenAI points at alongside data analysis [21], and it is worth nothing to somebody drafting two emails a day.
The five-times figure has an edge on it. According to runtimewire's reading of OpenAI's models and limits documentation, the extra included usage applies to ChatGPT Work and Codex, individual ChatGPT model allowances are counted separately, and the upgrade does not raise the limit for every model or feature [8]. Consumption is not denominated in prompts either: it moves with the model, the size of the task, the reasoning effort and the volume of work, and long cloud-based tasks draw down more than short local messages [9]. Allowances reset weekly, and either seat type can buy shared workspace credits once the included capacity is gone [6]. Nobody computes a break-even from that. What an admin can do is list the people who currently hit the ceiling and price them one at a time.
The billing terms all push the same way. New seats activate immediately with a prorated charge, reductions wait for the next renewal, and purchased seats are billed even when no member has been assigned to them [13]. A four-person trial is quick to start and takes a renewal cycle to unwind, so the seat count should follow observed behaviour rather than an expectation of it. Monthly billing costs $25 more per seat per month than the annual rate, which is $300 a year per seat, 25 percent more [18]. The cheaper rate is also the one that locks the seat in for the year.
There is a ceiling on the whole arrangement. Fill the 200-seat self-serve limit [12] with Premium at the annual rate and the workspace runs $20,000 a month, $240,000 a year [20], on a plan that does not accept invoices, purchase orders, bank transfers or ACH and does not include API usage [14]. Well below that number, a mixed-seat workspace stops being a self-serve convenience and becomes a contract that nobody negotiated. The product list is identical at both tiers [7], so the only lever left is how many names are on the expensive side of the ledger.
What to watch
- Whether OpenAI publishes included allowances in units an admin can forecast, instead of describing consumption as varying with model, task size and reasoning effort.
- Whether the rolling five-hour ceiling on Standard seats survives, once OpenAI can see how much Premium demand that single limit generated.
- Whether the 200-seat self-serve cap or the refusal of purchase orders and ACH moves, since a heavily Premium workspace already spends like a contracted account.