Anthropic and the GSA announced a 40% discount on Claude Enterprise, Claude for Government and the Claude API, with no minimum spend, now open to state, local, tribal and territorial buyers for the first time. Anthropic estimates about $1 billion in savings and says roughly 500 US government organizations already use Claude. The Justice Department filed a notice of appeal the previous day seeking to restore a federal ban on agency use of Anthropic models, and a different appeals court upheld the Pentagon’s “supply chain risk” label last month. Buyers get a real price cut, but the legal question of which agencies may use Claude at all is still open.
The Justice Department filed a notice of appeal on 26 October seeking to restore a federal ban on agency use of Anthropic’s models. The next day, the GSA and Anthropic announced a 40% discount across every Claude purchase surface for the same government. Both things are now true at once.
What the GSA agreement covers
The discount applies to Claude Enterprise, Claude for Government and the Claude API, with no minimum spend, whether a buyer purchases direct or through a cloud provider. Anthropic puts the estimated value at about $1 billion in savings. The newer part is eligibility: state, local, tribal and territorial governments can use the agreement for the first time, which is what “all levels of government” in the announcement refers to.
On adoption, Anthropic reports roughly 500 US government organizations using Claude, and an eightfold increase in Claude for Government weekly active users since February. Claude for Government reached general availability on 30 September, 28 days ago. The post also lays out the compliance footing agencies care about: FedRAMP authorization, plus deployment at DoD Impact Level 4 and Impact Level 5.
The security numbers Anthropic brought along
Two case studies in the announcement are worth more than the procurement details. In one, Claude Mythos caught a vulnerability that a human penetration test had missed. In the other, a ransomware response that Anthropic characterizes as a typical two-to-three-day job finished in eight minutes.
Treat the second figure carefully. The eight minutes is Anthropic’s, and so is the baseline it beats; no independent source establishes what a two-to-three-day ransomware response means in practice, or which parts of it the model actually performed. The first case study is the sturdier claim, and it fits a pattern we’ve been tracking: Anthropic’s models generating verified vulnerability findings faster than anyone fixes them, and building working exploit chains across five platforms. A model that beats a human pentest on the defensive side is the same capability pointed the other way.
Claude Mythos is the model behind Anthropic’s Project Glasswing cyber work. The government-facing pitch here is the defender half of a capability the company has also been publishing offensive results for.
Two parts of the federal government, two answers
The announcement’s framing works only for GSA’s civilian procurement posture. Elsewhere in the same government:
- A California federal judge ruled on 27 August that the Pentagon’s designation of Anthropic as a “supply chain risk” was unlawful.
- The D.C. Circuit, in a separate suit, upheld that same designation 2-1 on 25 September, 32 days before the GSA news.
- DOJ is now appealing the California ruling to the Ninth Circuit, which would reinstate the broader ban if it succeeds.
So a procurement agency is cutting prices and expanding eligibility while the Justice Department litigates to stop agencies buying at all, and one appeals court has already sided with the Pentagon against the company. The GSA deal describes what agencies may purchase today. It settles nothing about what they will be permitted to run in six months. Readers on Hacker News landed on the same gap: Claude remains a supply chain risk to the DoD while civilian adoption accelerates.
The discount is measured against list price, not against what agencies paid
In August 2025, GSA struck its original OneGov deal with Anthropic: Claude for Enterprise and Claude for Government offered to all three federal branches for $1 per agency for a year. Against that baseline, 40% off list is a price increase of several orders of magnitude.
That comparison is slightly unfair, since a $1 promotional year was never the steady state, and the new terms cover more buyers and more products. It’s still the relevant baseline for an agency budgeting next year. Secondary sources put the expiry of the $1 pricing at 31 October, which I could not confirm on a primary GSA page, so treat the exact date as unverified. The direction is clear either way: the free trial is ending and the real bill is arriving, discounted.
That is the normal shape of a land-and-expand deal, and GSA buyers were never going to get $1 forever. What makes this one odd is that the expansion and the litigation to shut it down are running in parallel, from the same government, in the same week.
Sources: Expanding our partnership with GSA to bring Claude to all levels of government • GSA Strikes OneGov Deal with Anthropic • DOJ appeals ruling lifting Anthropic ban
Coverage: Appeals court upholds Pentagon designation • Judge rules Pentagon label unlawful • Hacker News discussion
Related on this blog: Amazon Ran a Project 860% Over Budget and Took Five Months to Notice • [AUTO] The Shadow AI Risk Is Real, but the Playbook Is Conventional • [QT] Amazon’s AI Spending Problem