Meta places Llama on GSA's OneGov menu with no price to negotiate

An agreement where the pricing line is blank

On September 22, 2025, the GSA announced that Meta's Llama open source models are now part of its OneGov strategy, making them accessible across federal departments and agencies. OneGov exists to eliminate individual agency negotiations and reduce duplicative procurement effort.

The unusual part is stated plainly in Meta's own text: this arrangement required no procurement negotiations at all, because Llama is freely available. There was no price sheet, license fee, or usage tier to haggle over.

Unlike traditional OneGov agreements, this arrangement required no procurement negotiations because our Llama models are freely available.Montana Labs

That single sentence reframes what GSA actually delivered. This is not a bulk-purchase discount like a typical vendor OneGov deal. It is a distribution and endorsement channel for software that already costs nothing to obtain.

The work moved from contracts to verification

With no commercial terms to settle, GSA's contribution shifted to backend work — verifying that Llama meets federal requirements and providing what Meta describes as consistent, streamlined access across government.

That verification is the real deliverable. Any agency could already download Llama; what OneGov adds is a federal stamp that the models clear internal requirements, and a single reference point so each agency does not repeat the same evaluation independently.

Meta ties the move to concrete policy anchors: America's AI Action Plan, plus OMB Memoranda M-25-21 on accelerating federal AI use and M-25-22 on efficient AI acquisition. The M-25-22 alignment is telling — the most efficient acquisition, in this case, is one with no acquisition transaction to run.

Self-hosting is the automation lever

Meta emphasizes that Llama lets agencies retain full control over data processing and storage. For federal automation work, that control is the operative feature, not a footnote.

Because the weights are publicly available and can run inside an agency's own environment, teams can build mission-specific systems that touch sensitive data without routing it through an external API. That removes a common blocker for automating workflows involving records agencies are not permitted to send off-premises.

Meta also frames this as reducing dependency on closed providers and supporting reproducibility. For an agency automating a process it must audit and defend, running an inspectable model it hosts itself is a materially different posture than calling a black-box endpoint.

What a zero-cost listing does and doesn't remove

The specific implication of this arrangement: it clears the procurement and licensing hurdle, but it moves the entire cost and risk downstream to the agencies' technical teams.

Meta's own language is precise — federal teams can build, deploy, and scale for free. Free here means the model weights, not the operation. Standing up hosting, securing it, evaluating outputs, and maintaining deployments remain real engineering commitments that GSA's verification does not perform.

So the headline saving — no negotiation, no license fee — is genuine but narrow. The value Meta cites for taxpayers depends on whether agencies have the internal capacity to run open models responsibly. This deal removes the reason to say no; it does not supply the people who make it work.