Request for Information (RFI) for Artificial Intelligence (AI) Model Access Services

The attached Statement of Objectives (SOO) provides additional context for the requirement. The information below summarizes Treasury’s current thinking and is intended to facilitate informed and constructive industry feedback.

The primary objective of this acquisition is to establish direct contractual relationships with frontier AI and Large Language Model (LLM) developers in a manner that maximizes flexibility, speed, and adaptability. Treasury seeks an approach that enables timely access to rapidly evolving AI capabilities, including newly released or updated models and, where appropriate, associated offerings necessary to support secure and responsible use over the period of performance.

This requirement represents a new approach to contracting for AI capabilities. Rather than procuring a fixed set of tools or models through a one-time acquisition, Treasury is exploring a structure that provides sustained, direct access to frontier AI companies through a token-based procurement approach. This approach is intended to support mission-driven use while enabling the government to adapt to advances in model performance, safety, and functionality as the market evolves over the contract lifecycle.

2. ANTICIPATED CONTRACTUAL MODEL. Any award to support this requirement will likely be procured in the following manner:

• For purposes of this Notice, Treasury defines a "Frontier AI Company" as an organization that:
  • Develops and operates proprietary large language models (not exclusively  open-source implementations)
  • Demonstrates state-of-the-art or near state-of-the-art performance on recognized benchmarks (e.g., MMLU >75%, HumanEval >65%)
  • Maintains direct control over model development, deployment, and API access
  • Has established commercial API offerings with documented SLAs.
• Treasury is primarily interested in direct relationships with model developers rather than resellers, aggregators, or other intermediaries. However, Treasury may consider responses from other organizations that can demonstrate unique value in this space.
• Small businesses that partner with or resell frontier AI capabilities should clearly describe their value proposition and relationship structure with the OEM.
• Multiple Indefinite Delivery, Indefinite Quantity (IDIQ) contracts, or a similarly flexible contractual vehicle, directly with Frontier AI companies to support enterprise-wide access to frontier AI capabilities. The contract(s) will be available for use across the Treasury enterprise and by other Federal entities.
• The anticipated period of performance is one base year with up to four option years.
• Treasury intends to establish a standardized token measurement and pricing framework to enable:

• Consistent cost comparison across vendors
• Simplified budgeting and forecasting
• Transparent usage tracking and reporting
• Fair competition for task orders

Treasury intends to conduct a periodic refresh of contract holders during the life of the contract awards, potentially on an annual basis, to incorporate new vendors, models, and capabilities as the market evolves.

• Evaluation of the contracts (or orders) may use streamlined or alternative approaches, such as oral presentations, demonstrations, or comparative evaluations.
• Treasury is particularly interested in pricing approaches that reflect direct OEM engagement, high-volume government usage, and provides the government with the most favorable economics relative to standard commercial offerings.
• Orders will generally be issued in accordance with fair opportunity requirements; however, Treasury may limit competition or issue directed orders where permitted by applicable regulations and justified by the specific use case.

NOTE: There is no incumbent contractor currently supporting this effort.

3. APPLICABLE CODES

• NAICS: 518210 – Data Processing, Hosting, and Related Services
• PSC: DA10 – IT and Telecom – Business Application/Application Development Software as a Service (SaaS)

4. EVALUATION CRITERIA (ANTICIPATED). Quotes or submissions for any resulting solicitation may be evaluated based on some or all of the following criteria:

4.1. Token-Based AI Capability and Governance. Under such a factor, the government may consider:

• AI models available for token-based access and general capability scope;
• Methods for accessing and consuming tokens (e.g., APIs or managed interfaces);
• Approach to introducing new or updated models for token use over time;
• Data handling and governance practices related to prompts, outputs, and usage metadata, including ownership, retention, and deletion;
• Responsible AI practices, including bias mitigation, transparency, and explainability;
• Security architecture and compliance posture supporting token access;
• Availability of documentation and support related to token usage and account management;
• Identification of third-party services that process or access government data in connection with token usage;
• Exit and transition considerations, including termination of access and availability of usage records.

4.2. Cost Efficiency and Pricing Transparency.  Under such a factor, the government may consider:

• Clarity of usage-based pricing structures, including unit definitions, tiers, discounts, and volume considerations;
• Identification of all potential fees or charges;
• Flexibility of pricing to accommodate varying use cases and consumption patterns;
• Support for cost predictability and management over the period of performance.

4.3 Security and Compliance. Under such a factor, the government may consider:

• Current FedRAMP authorization status (or clear path to authorization)
• Data residency and sovereignty controls
• Encryption and access controls
• Audit logging and monitoring
• Incident response capabilities
• Continuous monitoring posture
• Supply chain security

4.4 Management Approach. Under such a factor, the government may consider:

• Service availability and reliability
• Support model and escalation procedures
• Performance monitoring and reporting
• Financial reporting accuracy and timeliness

5. INFORMATION TO INCLUDE IN A RESPONSE TO THIS NOTICE. Vendors responding to this notice are asked to provide the following:

• Company name;
• Unique Entity ID (UEI);
• Business size (and socio-economic status, if applicable);
• Existing government contract vehicles, if any, through which your company currently provides access to AI tokens;
• Questions, clarification needs, or recommendations to refine the requirement or contractual approach (limited to 30 questions and up to 25 recommendations);
• Identification of potential barriers to participation;
• Responses to the following question (limited to 1 page per question) – brevity and clarity are encouraged and tables and diagrams may be used within the specified page limit:

1. Are there any on-premises or cloud considerations, limitations, or dependencies Treasury should consider when ordering or using AI tokens, including data residency or tenant isolation considerations?
2. How does your organization define and measure “tokens,” and what variability exists across models, versions, or capabilities?
3. For representative government mission use cases (e.g., document analysis, code generation, conversational interaction, image generation), please describe typical proportional distribution of token usage across input tokens, output tokens, and any cache or context-related tokens.
4. Describe how government data (prompts, outputs, and usage metadata) is handled when tokens are used, including default retention, deletion, data export capabilities, and any use of such data for training, tuning, or analytics.
5. How are AI models made available for token-based access, how are model versions managed over time, and how are material changes or updates communicated to customers?
6. What security, authentication, and access control mechanisms support token usage (e.g., API keys, OAuth, audit logging, rate limiting, and revocation), and how are these controls applied to protect government data, including CUI?
7. What is your current authorization status (e.g., FedRAMP Moderate or Agency Authority to Operate), or what is your proposed path and timeline to obtain authorization, including your approach to continuous monitoring?
8. Describe the mechanisms available to support consumption of AI tokens, including API features, SDKs, documentation, authentication options, and any relevant usage constraints or rate limits.
9. What service availability commitments, maintenance practices, outage communications, and support models apply to token access, including escalation procedures for critical issues?
10. Describe your pricing structure for token-based access, including unit definitions, tiers, discounts, carryover or expiration policies, reporting granularity, and any tools or controls to monitor, cap, or budget usage.
11. What exit or transition considerations should the government be aware of when discontinuing token access or moving to another provider, including availability of usage records and data export formats?
12. What aspects of the proposed contractual model (IDIQ structure, periodic refresh, common token definition) may present challenges or risks, and what alternative approaches or refinements would you recommend?
13. Treasury recognizes vendors use different tokenization schemes. Treasury seeks industry input on the below areas. Note: This question does not require vendors to change their technical tokenization but rather provide clear conversion metrics for government planning:
    • Whether a standardized token equivalent is feasible
    • How to normalize across different tokenization approaches
    • Best practices for token measurement and reporting
    • Appropriate baseline for comparison (e.g., GPT tokens, character count)
14. To support consistent evaluation and normalization of token-based pricing, please describe your tokenization methodology and provide the approximate input token count for each of the representative text samples in the table below. In your responses, identify any model-specific variations, preprocessing assumptions, encoding differences, or other factors that may affect how tokens are calculated or billed. (see attached form)
15. What commercial indemnification terms are standard for AI model access (including IP infringement, data misuse, or model hallucination impacts)?
16. Given observed rapid price reductions in the AI market, what contractual mechanisms would best allow the Government to benefit from downward pricing trends over a 5-year performance period?
17. How are model retirements or deprecations handled commercially? What notice periods are typical, and how is customer continuity protected?
18. How does your organization manage capacity constraints or resource prioritization during periods of high demand, and what assurances can be provided regarding federal continuity and SLA adherence?
19. What tooling exists for enterprise-level policy enforcement (usage caps, content filtering, logging, per-bureau isolation, etc.)?

Vendors are kindly requested not to provide marketing materials as they are not useful for the purpose of this Notice and will not be reviewed.

6. HOW TO SUBMIT A RESPONSE TO THIS NOTICE. Responses to this Notice shall be:

• Submitted via email to [email protected] and [email protected];
• Use the subject line: “Treasury Notice – Frontier AI Company Tokens – February 2026”;
• Be provided in searchable Adobe PDF format (version 7.x or earlier), with at least 0.5-inch margins and no smaller than 10-point font;
• Be submitted no later than 9:00 AM Eastern Time on 23 February 2026. Responses received after this date and time may not be considered.

IMPORTANT NOTES:

• This notice is *NOT* a solicitation and does not obligate the Government to issue one or to award a contract.
• The government will not acknowledge receipt or provide feedback on submissions.
• Responses are not considered offers and cannot be accepted by the Government to form a binding agreement.
• The government will not reimburse any costs incurred in responding to this notice nor return any data provided.
• Submitted information may be used in development of future acquisition strategies or solicitations.
• Respondents should ensure all provided information provided is complete, accurate, and sufficiently detailed as it may be used for making acquisition approach decisions or better refine its requirements.
• The government may elect to engage further with some, all, or none of the respondents to this Notice.