OpenAI Offers Half-Price Licenses to All Levels of U.S. Government Without Fees
OpenAI has reached a new agreement with the U.S. General Services Administration (GSA): the ChatGPT license fee for federal, state, local, and tribal governments will be reduced from the usual $15 per user per month to zero, with eligible usage further discounted by 50%, and no minimum spend or seat commitment required. The agreement lasts for 27 months, from October 1 to December 31, 2028.
This follows last year's OneGov pilot program and is not an unlimited free offer. Last year, federal agencies could use the model for $1 per year, which the GSA called its deepest technology discount; that pilot program expires on September 30. The GSA stated that approximately 3.5 million federal employees were able to use ChatGPT during the pilot, saving about $1.4 billion in total with other AI agreements, and the overall savings for OneGov was about $1.68 billion. OpenAI claims the existing agreement covers over 1 million government employees, and the new agreement expands eligibility to about 23 million public sector workers in the U.S.
The procurement path follows OneGov, allowing direct orders, through dealers, or cloud marketplaces. It covers ChatGPT Enterprise, Codex, and related models, with a 50% discount for token billing in FedRAMP-authorized environments. The GSA has included its own AI clauses to govern government data. OpenAI also provides usage estimates, spending controls, onboarding, and FinOps guidance, and is expanding the Government Academy. In terms of cybersecurity, verified government entities can use the Daybreak Blue defense channel at half price, while Daybreak Red can be applied for at standard rates.
The pricing has changed from "one dollar per year" to "license free, usage at half price"; heavy usage agencies will see bills higher than the pilot year, while light users still face no barriers. Anthropic also negotiated a similar one dollar per year term after OpenAI's deal in 2025; three OneGov AI contracts expire simultaneously. Deployment options include ChatGPT FedRAMP, API FedRAMP, and ChatGPT Gov on Azure, along with the upcoming AWS GovCloud.
In market mechanics, this shifts from customer acquisition subsidies to usage-based revenue sharing. Buyers must decide whether to continue using the service after September 30, including newly included state, county, and tribal governments; the seller is OpenAI, exchanging zero seat fees for token revenue. Funds will flow from agency operating budgets to usage bills, no longer stuck in procurement seats. Beneficiaries are agencies that have developed usage habits and can incorporate half-price tokens into next year's budgets, as well as OpenAI entering state and county markets under GSA terms; those under pressure are departments still wanting to cap at one dollar, and competitors Anthropic and Google who must bid within the same procurement framework. The event is driven by the pilot's expiration, not a sudden emergence of new demand.
On a supplementary level, the 23 million refers to eligible populations, not activated seats. Actual spending depends on whether agencies turn on usage-based billing and which models are included at the 50% discount in FedRAMP environments.
Source: Public Information
ABAB AI Insight
One dollar per year treats the federal government as a showcase: first, let 3.5 million develop usage habits, then switch to half-price token billing at the contract expiration. Zero license fees maintain access, while half-price usage is the revenue switch. The GSA calls this OneGov 2.0, while OpenAI describes it as covering 23 million public sector workers. Both are the same thing: the pilot ends, and the government shifts from a symbolic buyer to a measurable consumption customer.
The capital path is "penetrate first, measure later." With the removal of seat fees, sales no longer negotiate per account, and billing directly follows token usage, with states, counties, and tribes included in the same discount table, saving 5,000 local tenders. Money flows from agency IT budgets to OpenAI's usage revenue, then to compliant deployments on Azure and AWS. The half-price cybersecurity products lock security teams into the same account. The motivation is to renew established workflows with zero barriers when Anthropic and Google’s contracts expire simultaneously.
This mirrors Microsoft's strategy of using free Office 365 for educational purposes to change generational habits, and AWS's credit subsidies for startups: both shift the price war from list prices to "pay after using." The industry phase shifts from expanding government procurement lists to controlling default models. Whoever becomes the default for agency documents, code, and cybersecurity will have conversion costs in the next round of negotiations by the end of 2028.
The structural change represents a transfer of pricing power. The mechanism is: once government procurement completes identity verification, marginal costs are based on tokens rather than licenses; writing licenses to zero shifts competition from "who has the cheaper annual fee" to "whose model is already integrated into the process." Half-price is not the end of discounts, but transforms the federal cloud from a sample room to a billing space.
ABAB News · Cognitive Law
One dollar per year buys habits; half-price tokens collect habits.
When licenses are free, the real contract lies in usage.
The public sector's default model will become the conversion cost in the next round of procurement.