GSA swaps OpenAI's dollar-a-year ChatGPT deal for half-price tokens

A dollar a year is not a budget line. It is a rounding error a CIO approves once and never revisits, and until the end of September, ChatGPT access costs federal agencies exactly that ^[10]. From Oct. 1 the price follows use: 50% off token-based consumption across ChatGPT models, with no platform-access fee, no minimum orders and no spend commitment ^[2][4].

No minimum takes the floor off the bill and the ceiling with it. The standard licensing fee OpenAI published Thursday is $15 per user per month ^[5], or $180 a year for one seat ^[1]. Applied to the more than one million government employees who already have access ^[9], a year at that list price would run at least $180 million ^[2]. Nextgov reported that the move to consumption pricing will likely mean higher costs for agencies, and that agencies will have to decide whether to keep offering employees access ^[11]. GSA's release does not say whether the 50% discount touches the seat fee at all; the discount is stated on token-based usage ^[2][4].

"As agencies increasingly integrate AI into their regular operations, providing consumption-based access is the next logical step," said Laura Stanton, acting commissioner of GSA's Federal Acquisition Service ^[3]. Twenty-seven months starting Oct. 1 covers two complete federal fiscal years and the first quarter of a third ^[4].

OpenAI expects the agreement to make roughly 23 million people eligible, counting federal, state, local and tribal agencies buying direct, through resellers or on supported cloud marketplaces ^[8][6]. That is about 22 million more than the population with access today ^[3]. Eligibility is a contract term and access is a provisioning decision; neither counts a task anyone actually ran. Consumption billing does count them. Under a flat fee, an idle seat and a daily-use seat cost the same; under tokens, the idle one bills nothing.

OpenAI also said "every verified government entity will be approved for our advanced cyber-defender Daybreak Blue access, all at 50% off standard commercial pricing, with scaled training and enablement to support adoption" ^[16]. Daybreak Red is available to government partners on request at standard commercial pricing ^[17].

Anthropic's Claude and Google's Gemini OneGov deals expire at the end of the month ^[12], and Nextgov reported it is unclear whether GSA will extend them ^[13]. Birgit Smeltzer, director of GSA's Office of IT Products, said at GovCIO Media & Research's Federal AI Forum last month that some of the original equipment manufacturers have already agreed to extend some of the limited time offers, and some are looking at producing new offers for the agency ^[14]. GSA says OneGov has saved agencies about $1.68 billion, with $1.4 billion of that from AI-related agreements ^[15], roughly 83% of the total ^[5].

Two questions on last month's data will settle most of this before Oct. 1: what share of provisioned seats produced any request, and how many requests came from the top tenth of the seats that did. Broad and shallow means the seats were a convenience, the consumption bill will be small, and the case for renewing every seat is weak. Narrow and deep means a handful of teams built real workflows, and the bill belongs on their budget rather than on a central IT line. Broad and deep means someone has to forecast a variable cost across two fiscal years ^[4]. If neither is true, the dollar-a-year price was the reason the deployment existed ^[10].