Half-Price Tokens for Washington: GSA and OpenAI Sign the OneGov 'AI 2.0' Consumption Deal

On September 10, 2026, the U.S. General Services Administration quietly executed one of the most consequential AI procurement moves of the year. GSA announced a new OneGov agreement with OpenAI — officially the first deal under the next phase of its OneGov AI strategy — that makes ChatGPT models available across government through discounted, consumption-based pricing. Inside government and on the procurement blogs, it is already being called the “AI 2.0” agreement. It is a 27-month offer expected to take effect October 1, 2026, giving agencies more than two years of stable, discounted pricing on the models that have become the de facto standard for AI-assisted government work.

What the agreement actually delivers

The structure of the deal marks a deliberate break from the earlier generation of OneGov AI offers. Three provisions matter most.

Discounted, consumption-based access. Participating governments receive a 50% discount on token-based usage across ChatGPT models, including those running in FedRAMP-authorized environments. Instead of flat per-agency pilot licenses, agencies now pay per token — at half the commercial rate — which aligns cost directly with usage.

No minimums, no lock-in. There is no platform-access fee, no minimum order, and no spend commitment. Agencies pay only for what they actually use. For small bureaus that were hesitant to commit budget during the pilot era, the barrier to entry has effectively vanished.

Broad eligibility and multiple ordering paths. The agreement covers federal (executive, legislative, and judicial), state, local, and tribal governments — a remarkably wide net. Ordering can happen directly, through resellers, or via supported cloud marketplaces. The deal bundles training and enablement resources for agency adoption and folds in GSA’s AI terms and conditions intended to safeguard government data.

GSA Administrator Edward C. Forst framed the agreement as positioning “the federal government for the future by integrating advanced, AI-enabled capabilities into agency operations.” Sam Altman’s statement leaned on service delivery: “Giving public servants secure access to the best AI tools can help government be more efficient, strengthen cybersecurity, and improve the services people rely on.”

From $1 experiments to half-price scale

To understand why this deal matters, you need the history. OneGov began as a vehicle for small-bore technology pilots — the famous $1-per-year ChatGPT agreements that let agencies try frontier AI before committing to a lengthy acquisition. As Laura Stanton, Acting Commissioner of GSA’s Federal Acquisition Service, put it: “We launched OneGov AI deals because agencies wanted the ability to try new technologies before committing to a lengthy acquisition. As agencies increasingly integrate AI into their regular operations, providing consumption-based access is the next logical step.”

The numbers behind that progression are striking. Since inception, OneGov has generated roughly $1.68 billion in cost savings for the federal government — of which approximately $1.4 billion stems specifically from AI agreements that expanded access to AI tools for roughly 3.5 million federal employees. Those original offers are now expiring, and GSA is renewing agreements where appropriate while shifting the model from experimentation to sustained consumption.

The economics tell the story of the shift: the earlier deals priced access as low as $0.47 to $1 per agency — essentially free trials at scale. The new agreement replaces that with metered usage at half price. For AI-saturated agencies, costs will scale with adoption; for cautious ones, the no-minimum structure means near-zero risk to start. OpenAI, for its part, trades headline-sized contract values for something arguably more valuable: a durable, governmentwide channel into public-sector workflows for over two years, across every branch and level of American government.

Policy alignment — and a competitive message

GSA explicitly tied the agreement to the White House’s America’s AI Action Plan and to OMB Memoranda M-25-21 (“Accelerating Federal Use of AI through Innovation, Governance, and Public Trust”) and M-25-22 (“Driving Efficient Acquisition of Artificial Intelligence in Government”). Agencies will be able to procure through GSA’s Multiple Award Schedule once the offer takes effect on October 1.

There is also a competitive signal embedded in Stanton’s remarks: “We welcome additional AI technology companies to engage with GSA and explore opportunities to bring their capabilities to agencies through OneGov as we continue to expand choice, increase competition, and deliver value for taxpayers.” OpenAI is the first mover in the consumption-based phase, but GSA is openly courting rivals — and prior OneGov AI deals already spanned Google, Anthropic, and xAI. The government clearly wants multiple frontier providers on similar terms, both to avoid single-vendor dependency and to keep pricing pressure alive.

The deeper context: buying AI like electricity

The “AI 2.0” agreement represents a maturation in how the U.S. government thinks about procuring intelligence itself. The first phase answered “can agencies use this?” The second answers “how does government buy something whose value is entirely a function of volume?” Consumption-based pricing — tokens as the metered unit — treats AI less like software and more like electricity: a utility whose cost scales with demand, and whose strategic value lies in reliability and price predictability over years, not quarters.

There are risks in that analogy. Metered pricing means agency AI budgets become a direct function of employee usage, and usage of coding and drafting assistants has a well-documented tendency to compound. A 27-month price floor protects against rate shocks but not against usage growth. Privacy advocates will also note that easier, cheaper, governmentwide access accelerates the absorption of AI into administrative functions — eligibility determinations, correspondence, analysis — faster than oversight mechanisms have historically moved, FedRAMP authorization and GSA’s AI terms notwithstanding.

For OpenAI, the deal lands amid a season of intense scrutiny — antitrust inquiries, Senate investigations, and public soul-searching from its own leadership about the pace of capability gains. A stable, multi-year public-sector channel with the explicit endorsement of GSA’s leadership is a counterweight to that narrative: it positions the company as infrastructure for the American state rather than merely a fast-moving consumer lab.

Either way, October 1 is the date to watch. When the agreement takes effect, roughly 3.5 million federal employees — plus state, local, and tribal governments — get frontier AI at half price, with nothing down. The era of experimenting with AI in government is over; the era of running on it has a contract.