GSA Sets October Shift to Zero-Fee, Usage-Based OpenAI Pricing

GSA and OpenAI have scheduled a OneGov agreement to begin October 1, replacing a $1-per-agency ChatGPT Enterprise promotion with a $0 platform fee and 50% off eligible usage. It is not free AI: agencies will place and fund their own orders, and the public materials reviewed through September 10 did not contain the detailed rate card needed to calculate particular token bills.

Price without the seat charge

The scheduled agreement runs for 27 months, from October 1, 2026 through December 31, 2028. OpenAI says it eliminates the standard $15-per-user monthly platform or license fee, has no minimum commitment, and applies a 50% discount to eligible usage. The core published pricing basis is token consumption across eligible ChatGPT models, including FedRAMP-authorized environments.

That distinction between access and use is the deal’s central financial change. A government office can add ChatGPT Enterprise without a recurring seat charge, but its spending still rises with eligible model use. The authenticated OneGov term sheet and full ordering sheet were not publicly available at the September 10 cutoff, leaving model-specific rates, token categories, overage formulas, and marketplace coverage unverified.

The outgoing offer already had limits

The earlier promotion was not a conventional per-seat deal. It charged $1 for Core ChatGPT Enterprise per participating agency, not per user or token. It also included 100 million Advanced Model Units for its first 60 days, while additional Advanced Model Unit credit packs and support packages carried separate tiered prices.

So the October plan is not simply a switch from unlimited access to metering. Both arrangements separate a low entry price from some form of additional consumption or service cost. The change is that the successor makes discounted token-based usage the stated core cost structure, while removing the platform fee altogether.

The published calendar creates a direct handoff. The promotional agreement expires September 30, and the successor is scheduled for the next day. As of the evidence cutoff, though, the new agreement was announced rather than active—an important difference for agencies assessing what they can order now versus what is expected to become available in October.

A wider public-sector buyer pool

The new arrangement explicitly reaches federal executive, legislative, and judicial organizations, as well as state, local, and tribal governments. That is a clear expansion beyond the old offer’s federal focus. Public descriptions of the previous program differed on whether its federal eligibility was limited to the executive branch or extended across federal agencies and bodies, but none extended it to state, local, or tribal governments.

The broadened eligibility does not mean every buyer receives the same product route. OpenAI markets the program around ChatGPT Enterprise, Codex, and OpenAI models. Its government intake form asks organizations to identify interest in ChatGPT FedRAMP, direct API access, AWS Bedrock API access, and cyber or specialty models. Those choices show the potential procurement paths being considered, not final confirmation that every listed option has identical coverage or pricing.

The agreement’s published access points

• ChatGPT Enterprise, Codex, and OpenAI models are named as program offerings.
• ChatGPT FedRAMP and direct API access appear on OpenAI’s government intake form.
• AWS Bedrock API access and cyber or specialty models also appear as stated interests.

A documented exception

The advertised 50% discount is not a universal price for every cyber offering. OpenAI says every verified government entity will be approved for Daybreak Blue at 50% off standard commercial pricing. But an agency that requests Daybreak Red for advanced vulnerability research, exploit validation, and red teaming is offered that access at standard commercial pricing.

That exception narrows a tempting interpretation of the headline discount. Agencies considering cyber-defense work have to distinguish between the program’s general eligible-usage terms and the treatment of a specific, more advanced product. The difference is especially relevant because the agreement pairs general access with an expanded public-sector audience, rather than presenting a single flat price for the full catalog.

The budget responsibility stays local

GSA’s negotiated terms do not create a centrally funded pool of OpenAI use. The program is agency-funded: each agency places and funds its own order. The successor also removes minimum order and spend commitments, but agencies still must follow applicable FAR 8.4 ordering, competition, and documentation procedures.

That changes the operational question for a public buyer. The old offer made entry unusually inexpensive at the agency level. Under the planned successor, an office can start without a platform charge or a required spending floor, but it still needs a funded order and a way to keep usage within its own budget. The relevant decision is less about purchasing a fixed number of licenses than estimating how much model consumption a particular workload will generate.