Claude for Government goes live as $1 federal AI deals wind down

Anthropic made Claude for Government generally available to US federal and state agencies on Wednesday, September 30, after a public beta that began in July. The service delivers Claude’s coding and agent features through a FedRAMP High authorised environment. FedRAMP is the US government’s security approval for cloud services, and High is its strictest level for unclassified data.

The bigger change is how agencies pay. “No seat fees,” Anthropic says. Agencies buy usage in fixed increments “with a hard not-to-exceed cap, so spend does not exceed what an agency has obligated”.

The $1 era is ending

That pricing arrives at a turning point for federal AI deals. Under the General Services Administration’s OneGov programme, the big labs offered their tools to agencies for token sums, and those offers were due to expire at the end of September. Nextgov/FCW reports that Anthropic extended $1-per-user access to Claude to October 31, Google extended Gemini at $0.47 per agency per year to November 15, and OpenAI replaced its $1-per-agency price with a new deal, starting October 1, that gives agencies a 50% usage-based discount on ChatGPT models.

“As those offers expire, GSA is renewing agreements where appropriate, ensuring continued access to critical AI tools, and delivering the best value for agencies and taxpayers,” GSA said in a statement quoted by Washington Technology.

So what changes for an agency? Under a near-free deal, the only question was who gets access. Under metered billing, the question becomes how much work runs through the model, and whose budget pays for it. Two of the three big suppliers are now steering agencies towards usage-based pricing. Anthropic’s announcement does not give unit prices for what comes after October 31.

Controls written for auditors

Much of the announcement reads as if it were written for the people who sign off on government systems. Administrators set spending and model limits for each group of users, track use by person and by model, and get alerts before a balance runs low. “Conversation history stays local on the agency-managed device,” and usage exports contain only metering data, so agencies can answer ATO and IG requests without moving sensitive material. An ATO is an authority to operate, the formal approval a government system needs; an IG is an inspector general, an agency’s internal watchdog.

Anthropic also says “sensitive operations on Anthropic’s side require two-person approval”, and administrative actions are recorded in an audit log. Two more products are arriving behind the desktop app: the Claude Code command-line tool and Claude for Microsoft 365, both rolling out in early access in the same environment. For agencies sitting on old software that needs rewriting, the coding tool may matter more than the chat window.

One department is still closed

The general release does not reach the military. The Department of War, as the administration now styles the Pentagon, decided that Claude “presents a significant supply chain risk” and excluded it from its supply chain under the Federal Acquisition Supply Chain Security Act. On September 25, the US Court of Appeals for the DC Circuit denied Anthropic’s petitions for review. Civilian and state agencies can buy Claude; the Pentagon will not.

What to watch

The first test is October 31, when Anthropic’s $1 offer ends. Will GSA sign a metered OneGov deal with Anthropic, as it did with OpenAI, and at what discount? Google’s turn comes on November 15. After that, watch how quickly the Claude Code tool leaves early access, and whether agencies publish what they actually spend once the meter is running.