Anthropic expands GSA partnership with 40% markdown on Claude products

Anthropic just handed the US government a coupon. Under an expanded OneGov agreement with the General Services Administration, public agencies now get a 40% discount on token usage across all Claude models.

The GSA announced the expansion on October 27, 2026. It replaces a symbolic price tag with something closer to a real utility bill, one that grows with how much agencies actually use the AI.

What the new deal covers

The headline change is the pricing structure. Instead of a flat fee, agencies now pay based on consumption, measured in tokens.

Under the new terms, every one of those tokens costs 40% less than standard rates. The discount applies to all Claude models, not a select few.

The scope is broad. The agreement extends to every federal agency across the executive, legislative, and judicial branches.

It also reaches beyond Washington. State, local, and tribal governments are eligible too, which widens the potential customer base considerably.

Procurement has been kept deliberately simple. Agencies can buy directly or go through resellers, and there is no minimum order requirement.

The deal also opens access to authorized government environments, including those that require FedRAMP compliance for sensitive unclassified work. FedRAMP is the federal security standard cloud services must meet before agencies can trust them with government data.

From a $1 pilot to a real pricing model

This agreement builds on an earlier arrangement that started in August 2025. Back then, Anthropic offered Claude to government users for a nominal fee of $1 per user or agency.

That earlier offer supported FedRAMP High workloads. It was later extended until October 31, 2026, just days after the new pricing was announced.

The shift to consumption pricing suggests the open house phase is winding down. The research describes it as part of a transition from evaluation to practical operational deployment.

The move fits within the GSA’s broader OneGov strategy. That program has reportedly saved the federal government more than $1.8 billion, with approximately $1.6 billion of those savings tied to AI-related contracts.

OpenAI’s steeper offer sets the bar

Anthropic is not the only AI lab negotiating with the GSA. A parallel agreement with OpenAI, announced in September 2026, offers a 50% discount on token use through December 2028.

That puts Anthropic’s 40% markdown in a slightly awkward spot. On paper, its rival is cutting a deeper deal and locking it in for longer.

Still, raw discount percentages do not tell the whole story. What an agency actually pays depends on the base rates, which model it chooses, and how many tokens a given task consumes.

What this means for government AI adoption

For agencies, the most immediate impact is lower friction. No minimum orders, multiple buying paths, and FedRAMP-compliant access remove several common reasons to delay adoption.

The consumption model also changes how budgets work. Agencies pay for what they use, which makes small, targeted deployments easier to justify than large upfront commitments.

For the AI industry, the GSA is effectively running a pricing competition in public. When OpenAI offers 50% and Anthropic offers 40%, every other vendor chasing government contracts now has visible benchmarks to match or beat.

The inclusion of state, local, and tribal governments is worth watching closely. That segment is fragmented, with thousands of potential buyers, and a federal pricing vehicle could make it far easier to reach.

Another thing to monitor is what happens as the original $1 arrangement expires on October 31, 2026. Agencies that onboarded during the nominal-fee period will now face real, if discounted, costs.

Finally, the side-by-side deals with Anthropic and OpenAI suggest the GSA is avoiding dependence on a single AI provider. Keeping multiple vendors on discounted terms gives the government leverage in future negotiations.