The GSA OneGov ChatGPT Deal Ends the $1 Era and Replaces It With a Meter

What the GSA OneGov ChatGPT agreement changes on October 1

GSA and OpenAI announced a new OneGov agreement on September 10, 2026. It takes effect October 1, 2026 and runs 27 months, through December 31, 2028. The GSA OneGov ChatGPT terms are consumption-based. Agencies get 50% off token usage across ChatGPT models. There is no platform-access fee, no minimum order, and no spend commitment. All three federal branches are eligible. So are state, local, and tribal governments.

If you sell software or services to federal agencies, this is a pricing event, not a new pipeline. Your buyer’s default comparison through 2028 is frontier AI at zero platform fee and half-price usage. Every build-versus-buy pitch you make has to beat that baseline on October 1. Schedule holders have a narrower stake. The agreement names resellers as an ordering channel.

Three things separate this from the deal it replaces:

• The prior $1-per-agency offer reached the executive branch. The new one covers all three branches plus state, local, and tribal governments.
• Price moved from a flat nominal fee to a metered rate, so cost now scales with use.
• Term moved from one year to 27 months, long enough to plan a multi-year program around.

Three of GSA’s AI offers expire September 30, and only one has a successor

OpenAI’s $1-per-agency ChatGPT Enterprise offer expires September 30, 2026. So does Anthropic’s $1-per-user Claude offer. So does Google’s $0.47-per-agency Gemini for Government offer. Those are nominal negotiated unit prices, not obligations and not appropriations. As of September 16, only OpenAI has posted successor pricing.

That table is narrower than “OpenAI won.” OpenAI has a priced, metered, multi-year successor to a promotion. That makes it the first OneGov AI offer to graduate from a loss leader into a rate card.

I would treat September 30 as a hard deadline for competing software vendors. Until then you argue price against a nominal fee. After that, against a published rate.

Consumption pricing moves budget risk onto the ordering agency

A flat $1 fee cannot surprise a buyer. A meter can. No platform fee, no minimum order, and no spend commitment removes the floor on agency cost. It also removes the ceiling. An agency that scales ChatGPT across its workforce now carries a variable IT line. That line behaves like cloud consumption, not like a license.

The sequence an agency follows:

1. Confirm eligibility. It now extends past the executive branch to legislative and judicial agencies, and to state, local, and tribal governments.
2. Pick an ordering channel: direct purchase on the Multiple Award Schedule, an authorized reseller, or a supported cloud marketplace.
3. Deploy into a FedRAMP-authorized environment. The agreement incorporates GSA’s AI terms and conditions to safeguard data.
4. Estimate usage and set spend controls before scaling. Nothing in the pricing caps monthly cost.
5. Track consumption against the obligated amount on the order.

The acquisition vehicle here is the Multiple Award Schedule, not a new contract. Three consequences follow:

• An agency with schedule ordering authority can buy without running a new competition.
• A reseller already on schedule can take the order without standing up a vehicle of its own.
• Terms are set centrally, so an individual agency has little room to negotiate further.

Spend control, step 4 above, is where contractor-relevant work sits. OpenAI will supply buyer guidance, usage estimates, spend controls, onboarding, and financial operations guidance alongside the agreement. A vendor that covers those itself leaves a third party little to sell. I would not build a service line on it.

Competitive Landscape: who else holds a federal AI offer, and whose standing is contested

Only one of the five vendors on GSA’s OneGov schedule for AI in government has posted a successor. A second had its federal standing contested in court earlier this year.

• OpenAI holds the only successor agreement running through 2028. It prices metered usage at 50% off and waives the $15-per-user monthly commercial license fee. That $15 is foregone list price, not government spending.
• Anthropic’s Claude offer expires September 30, 2026. Its federal standing has been contested since February 27, 2026. GSA removed Anthropic from USAi.gov and the Multiple Award Schedule that day, following a presidential directive to stop using the company’s technology. A federal district court issued a preliminary injunction on March 26, 2026. GSA withdrew the removal on April 3, 2026 and restored Anthropic’s prior status. The company’s standing on the schedule is a separate and unsettled question.
• Google’s Gemini for Government offer expires September 30, 2026; its separate Workspace discount runs to December 31, 2026.
• xAI and Perplexity hold smaller offers into March and April 2027, so the promotional model is not finished.
• Carahsoft is the reseller contact GSA names for the Claude, Gemini, ChatGPT, and Grok offers. OpenAI runs its own partner network. Enrollment in it does not by itself make a firm an authorized reseller for ordering under the OneGov agreement. For that, the near-term path is a subcontract with a reseller already carrying the offers.

For a GovTech investor holding a federal AI software position, the floor matters more than the discount. Every portfolio company selling general-purpose AI capability into agencies now prices above a published $0-platform-fee baseline through 2028. That is a pricing-structure change. It says nothing about product quality. Anthropic and Google have until September 30 to post successor terms. Benchmark your own government pricing against the OpenAI rate until they do.

Re-pricing a product against a zero-platform-fee baseline

If you run a FedRAMP-pursuing SaaS or AI product company, your differentiation argument changes on October 1. A platform fee for general-purpose chat now competes against $0. Your buyer gets that capability without one. You can still charge for mission-specific data, integration into systems the agency already runs, and accreditation depth. Delivery accountability is a fourth. A self-service meter does not provide it.

• Founders and CEOs: name the one capability in your product that survives a side-by-side against general-purpose chat. Lead your next government pitch with it.
• Capture and BD leads: find every open pursuit that claims general-purpose language, summarization, or drafting as a discriminator. Rewrite those sections before your next submission.
• Schedule holders and systems integrators: ask your own schedule contracting officer whether your existing SINs already cover AI software resale. If not, open a channel conversation with Carahsoft this month.
• Agency technology leaders: before signing a separate agency-specific AI license, compare its terms against 50% off usage with no platform fee and no minimum.

Where the contractor work is, and what would turn it into a pursuit

The services layer around an agency ChatGPT rollout is a requirements signal today, not yet a validated federal contracting opportunity. Integration, training, change management, and security assessment are a separate claim from the agreement itself. No solicitation, sources-sought notice, or forecast entry supports them yet. The agreement underneath is real. An identified buying population, a named vehicle, and a named product make it a validated federal acquisition mechanism. The spending happens at order time.

For most readers I would hold position and reprice this quarter, rather than chase a capture. For a schedule holder with an existing channel relationship, the move is a teaming conversation this month. Agency buying patterns are still unformed. Treat that channel as unproven. No agency order under either agreement is publicly recorded. The $1 orders sat below the micro-purchase reporting threshold anyway.

If your plan was to undersell frontier chat on unit price, you have no play here through 2028. A $1M-$100M federal software firm does not beat that baseline on unit economics. Two events would turn this into a pursuit. One is an agency posting a requirement for AI adoption support with its own scope and dollars attached. The other is a recorded agency order under the new agreement. Low recorded consumption through 2027 takes it off the list. That would mean agencies took the price and never scaled use.

FAQ

Which budget line does an order under this agreement hit?

The ordering agency’s own operations and IT accounts. No appropriation or program line is dedicated to this agreement. GSA negotiated the price. Each agency funds its own order. An AI order competes for the same money that pays your contract.

Does a schedule holder need a new Special Item Number to resell under this agreement?

Ordering runs on the Multiple Award Schedule, with resellers as an ordering channel. No Special Item Number is designated for these offers. Your own schedule contracting officer can answer that against your existing SIN list.

How does a reseller make money on a zero-platform-fee, metered offer?

The margin does not come from the license spread. There is no platform fee to mark up, and the usage rate is set centrally. Margin has to come from what surrounds the order: onboarding, integration, spend governance, and accreditation support.

What would be the first observable evidence that agencies are scaling use?

A recorded agency order or obligation after October 1, followed by consumption or spend data from any ordering agency. Agency AI use-case inventories are the second place it would surface. Neither exists yet, because ordering does not open until October 1, 2026.

Which contracting shops matter if this turns into services work?

The ordering agency’s own contracting office, not GSA’s OneGov office. OneGov negotiates price and does not buy on any agency’s behalf. A services requirement tied to an AI rollout would come from that agency’s own acquisition organization and budget.

What evidence would change this call?

A successor OneGov offer from Anthropic or Google on comparable terms. That would remove the single-successor distinction. Low recorded consumption through 2027 would undercut the premise that agencies scale use rather than hold cheap access.