What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Federal agencies should define the public result they need, how they will measure it, and what counts as acceptable performance before choosing how to pay for AI. Token-based pricing can provide flexible access to changing models, but usage alone does not show that a service is accurate, reliable, or useful to a government mission.
What does it mean to buy AI by the token?
With consumption-based pricing, an agency pays according to measured usage, such as the tokens processed by an AI service. This can make it easier to start small and pay for actual use rather than commit to a large fixed purchase. It also makes cost depend on usage, so buyers need visibility into consumption and controls to prevent spending from outrunning the value delivered.
A September 10, 2026, General Services Administration announcement illustrates this model. GSA said its OneGov agreement for OpenAI would offer a 50% discount on token usage, no platform-access fee, no minimum orders, and no spend commitment. The 27-month offer was expected to begin October 1, 2026. These are terms of that specific agreement, not standard terms across AI procurement. GSA’s announcement also reported approximately $1.68 billion in OneGov savings, approximately $1.4 billion attributed to AI agreements, and expanded access to roughly 3.5 million federal employees. Those are GSA’s reported figures, not independently audited figures in the release.
Usage pricing answers how much service an agency consumed. It does not, by itself, answer whether that service improved a process or met a mission need. A discount can reduce the cost per unit while leaving the agency without a clear way to judge the quality or public value of the work.
#1 Best Overall
What should agencies buy instead?
Agencies should specify a result and a way to verify it, then select a payment structure that supports delivery and accountability. The result might be a completed deliverable, a service level, or a measurable improvement in a process. The right measure depends on the task: a count of AI-generated responses, for example, would be a poor substitute for checking whether those responses are correct and usable.
The U.S. Government Accountability Office recommends that agencies consider performance-based acquisition techniques for AI, including statements of objectives, performance work statements, quality assurance surveillance plans, and contract incentives. It also recommends attention to transparent pricing, knowledge transfer, data and model portability, and clear licensing terms. These techniques help buyers define what they expect and how they will assess performance; they do not establish that performance-based contracts are always cheaper or better than consumption-priced ones. GAO’s April 13, 2026, report identifies challenges with performance monitoring, data rights, competition, model reliability over time, and understanding AI-related costs.
Rank #2
How do the payment models differ?
| Decision point | Consumption-based pricing | Performance-based approach |
|---|---|---|
| Payment unit | Tokens or another measure of service usage | Accepted deliverables, service levels, or mission outcomes |
| What the measure shows | How much service was used; it does not establish mission value on its own | Whether specified acceptance criteria or outcomes were met, if they can be defined and verified |
| Cost control | Usage limits, consumption reports, and transparent prices | Milestones, incentives, or service-level controls, alongside oversight and integration costs |
| Adaptability | Can support flexible access as models and capabilities change; continuity and transition still matter | Can tie payment to defined performance; requirements and measures may need adjustment as capabilities change |
| Government control | Depends on contract terms for data rights, portability, licensing, and knowledge transfer | Can include those terms and access to evidence needed to evaluate performance |
| Risk allocation | Usage charges can accrue even when quality or mission value disappoints | Contract terms can allocate consequences for missed criteria, but measurement and attribution must be workable |
This is not a choice between a universally good and bad model. A consumption agreement with no minimum spend can be useful for controlled experimentation or variable demand. A performance-based structure can strengthen accountability where results can be stated and assessed. Agencies should compare the full costs and risks—including evaluation, integration, oversight, and supplier transition—not just the price unit.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →How should an agency design an AI purchase?
- Start with the mission problem. Describe the task, the people affected, and the result the agency needs before naming a model or supplier. GSA’s buyer guidance asks, “What tasks or processes would benefit most from AI assistance?”
- Set a baseline and acceptance criteria. Define how the existing process performs and what evidence will show that an AI-assisted process is acceptable. Include a quality assurance approach so the agency can detect errors and assess results rather than relying on activity counts alone.
- Pilot before scaling. Test with a sandbox or small user group. Assess accuracy, reliability, security, user impact, and cost in the intended setting before expanding use.
- Make costs visible and controllable. For usage-priced services, require clear pricing, consumption reports, and spending limits appropriate to the use case. Account for the cost of integration and oversight when comparing options.
- Protect the ability to change course. Address data rights, model and data portability, licensing, knowledge transfer, and transition or exit plans. These terms affect whether the agency can retain control and move to another approach if performance or needs change.
- Coordinate the right officials. Involve acquisition, IT, AI, data, security, and privacy officials, and confirm the authorization requirements that apply to the service. GSA’s buyer guidance says cloud service providers must be FedRAMP-authorized or in the process of obtaining authorization, and advises agencies to coordinate with their security staff.
- Monitor performance throughout the contract. Reassess quality, reliability, costs, and contract outcomes over time. Model capabilities and risks can change, so initial acceptance is not a substitute for ongoing evaluation.
These steps synthesize GAO recommendations and GSA buyer guidance; they are not a claim that every clause or control is mandatory for every procurement. GSA’s Buy AI guidance also lists procurement channels and offerings, including USAi, OneGov agreements, cloud services, and GSA contract vehicles. Agencies should check the live guidance for current purchasing information rather than rely on older price listings.
Rank #3
What does the government’s own example show?
GSA’s 2025 AI use-case register describes a pre-deployment, experimental retrieval-augmented generation (RAG) assistant intended to coach acquisition staff as they draft outcome-based requirements. The stated goal is feedback comparable in accuracy and reliability to that of experienced human facilitators. That is an aim under evaluation, not evidence that the assistant has achieved the goal in production. The GSA use-case register illustrates the distinction between defining an intended outcome and demonstrating that a system delivers it.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Why does the distinction matter now?
GAO reports that federal agencies more than doubled their use of AI from 2023 to 2024. It also reports that investment in the AI sector grew more than thirteenfold since 2014. Those figures describe adoption and industry context, not procurement effectiveness. As use expands, agencies need a way to determine whether a particular purchase meets its intended purpose and what happens when it does not.
The practical standard is therefore not “never pay by the token.” It is to avoid treating token volume or a low unit price as proof of success. Define the mission result and evaluation method first; then choose consumption pricing, performance-based terms, or a combination that gives the agency useful flexibility without surrendering cost visibility or accountability.
Recommended Free Tools
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
