OpenAI is offering eligible U.S. government bodies a $0 user license for ChatGPT and a 50% discount on token-based usage. It sounds like a price cut. It is also a change in what agencies must learn to buy.
The General Services Administration announced the 27-month OneGov 2.0 agreement on September 10. It is due to run from October 1, 2026, through December 31, 2028, covering federal, state, local, and tribal governments. There is no platform-access fee, minimum order, or spending commitment, according to GSA. Ordering begins through the Multiple Award Schedule once the offer takes effect.
That is a meaningful opening for an agency that has been stuck between a small pilot and an enterprise contract. But it is not the same as free AI. OpenAI’s own announcement says the normal $15-per-user monthly license is waived while usage remains metered. Neither announcement publishes the model-by-model token rates, context limits, or the practical cost of agentic workflows that can call a model repeatedly. A 50% discount is easy to understand; a final invoice is not, until an agency knows what it will consume.
The procurement unit is changing
Seat pricing made the first question familiar: how many people need access? Consumption pricing adds harder ones: which model may each workload use, how many tokens can a department spend, how long are logs retained, and who can turn off an automated loop?
GSA says the agreement includes token-based access across ChatGPT models, including FedRAMP-authorized environments. It also says agencies can order directly, through resellers, or via supported cloud marketplaces. Those choices can matter to invoices, support, and contractual accountability; the release does not present them as identical buying experiences.
| Area | Publicly stated | Still material for an agency |
|---|---|---|
| Access | $0 monthly user license; no platform fee, minimum order, or spend commitment. | Which services, roles, and environments qualify. |
| Usage | 50% off token-based ChatGPT usage. | Published rate card by model, input/output token, context length, and tool call. |
| Security | FedRAMP-authorized environments are included; GSA terms are intended to safeguard data. | Data classification, retention, logging, agency configuration, and incident obligations. |
| Cyber defense | Verified government entities can access Daybreak Blue at 50% off standard commercial pricing, with training. | Eligibility, allowed use, workflow controls, and the separate terms for Daybreak Red. |
| Duration | October 1, 2026–December 31, 2028. | Price-change, portability, and transition provisions before the period ends. |
Source: GSA and OpenAI announcements, September 10, 2026. “Still material” is AI News Fab analysis of information not published in the announcements.
What the discount does—and does not—solve
For a small city or agency trial, no minimum spend is a real hedge. An organization with low, controlled usage can keep its direct software bill low. The free license also removes the awkward step of deciding which staff deserve a paid seat before a workflow has proved useful.
The trade-off arrives when a pilot becomes a service. A document assistant that processes a predictable batch is comparatively easy to budget. An agent that re-reads files, calls tools, retries errors, and serves many staff is a usage-management problem. Human review, prompt standards, permissions, audit logs, training, and export testing are not token charges, but they are part of the cost of operating responsibly.
OpenAI says ChatGPT Enterprise does not use business inputs or outputs to train or improve its models. That is an important stated protection, not a complete deployment design. Agencies still need to confirm the eligible service, chosen environment, data categories, retention settings, access roles, and procurement terms. “Enterprise” is not a substitute for a documented data boundary.
Cyber access creates a second decision
OneGov 2.0 also pairs the commercial offer with cyber-defense access. OpenAI says every verified government entity will be approved for Daybreak Blue at half the standard commercial price, with scaled training and enablement. It says Daybreak Red, intended for advanced vulnerability research, exploit validation, and red teaming, can be requested at standard commercial pricing.
That split is sensible, but it makes governance more—not less—important. Faster analysis can help a public-sector security team sort code, suspicious activity, or remediation options. A cyber tool also needs clear authorization, validation, and escalation rules. The announcements do not publish the approval process, technical limits, or data-handling detail needed to compare Daybreak with another supplier on like-for-like terms.
Competition will be decided by the paperwork
GSA says OneGov is intended to increase choice and competition, and the agency points to roughly $1.68 billion in OneGov savings since the program began, including about $1.4 billion attributed to AI agreements. Those are GSA program figures, not a measured result for this OpenAI contract.
OpenAI’s offer is strong at the front door: waived license fees, half-price consumption, broad eligibility, and a fixed 27-month availability period. Rival offers will need to be compared on the less visible columns: model-specific consumption rates, FedRAMP availability, contractual data use, administrative controls, procurement channels, and a credible route to export prompts and workflows. A lower token price can lose its advantage if the migration and governance costs are higher.
The first useful question is a budget cap
For agencies, the immediate move is not to pick a model. It is to build a small workload ledger: expected requests, average input and output size, permitted tools, owner, data class, human-review point, monthly ceiling, and shutdown condition. Then ask each supplier to price that same workload under the same assumptions.
OpenAI has scheduled a September 14 briefing for government leaders. The most valuable follow-up would be specific: a usable rate card, discount eligibility by service and model, how spend controls work in procurement, and what happens when an agency needs to leave. Until then, OneGov 2.0 is an unusually generous way to start. It is not enough information to declare a 27-month deployment cheap.