Claude Gets a 40% Government Discount Through GSA OneGov

If your government AI project is budgeted per seat, today's news changes the spreadsheet. On October 27, 2026, Anthropic and the US General Services Administration (GSA) expanded their OneGov agreement, and the headline is a Claude government discount of 40% on consumption usage. Below: what the deal says, what the sources leave open, and what a project manager should change before the old $1 pricing lapses on October 31.

What the Claude government discount covers

Per Anthropic's announcement and Unite.AI's summary:

• 40% off consumption usage. That means token pricing, where you pay for the text the model reads and writes. It covers Claude Enterprise, Claude for Government and the Claude API. Anthropic says this is the first time the discount spans every product surface.
• No minimum spend. Agencies pay for what they use.
• Wider eligibility. Federal buyers, plus state, local, tribal and territorial governments.
• Anthropic's estimate: the discounts will save the government about $1 billion.

From $1 per user to pay-per-token

The previous OneGov offer was $1 per user. GSA's OneGov listing still showed only that offer when I read it today, tagged as expiring October 31, 2026. Nextgov reported that it had already been pushed back a month from September 30.

The new deal swaps a flat seat price for a usage price. Nextgov notes OpenAI's OneGov deal went the same way, with a 50% token discount.

Seat prices are predictable. Token bills are not. A Forrester analyst told FedScoop that a small group of power users can drive a disproportionate share of costs.

How to buy at the OneGov price

• GSA Multiple Award Schedule (MAS): available today.
• Direct from Anthropic's Public Sector team: available today.
• Cloud providers: Anthropic says discounted purchasing through them arrives "in the coming months", and names Amazon Bedrock and Google Cloud Vertex AI. Anthropic lists both at FedRAMP High.

Unite.AI reports GSA saying ordering on major cloud platforms and through resellers is already open. Anthropic's own page says months away. Get the answer in writing from your contracting officer before you plan around either.

What the sources leave open

• Term. No source states how long the expanded deal runs. For comparison, Nextgov reports OpenAI's token-based OneGov agreement runs 27 months.
• Rates. Coverage gives no per-token prices. The 40% applies to whatever price sheet your contract uses.
• Authorization. Anthropic says Claude for Government has been generally available since September 30 at FedRAMP High. Claude Enterprise is still in process for FedRAMP Moderate. A discount is irrelevant if your data needs an authorization the product lacks.
• Partner program. Anthropic promises a public sector implementation program, with details "in the coming weeks".

Rebuild your budget as a usage forecast

On Monday morning, do four things:

1. Find out which licence each team is on and when it ends. Anything on the $1 offer needs a plan for November 1.
2. Replace seat counts with a token forecast per workload.
3. Add a growth multiple and a spend cap to the plan.
4. Choose a buying route you can use today (MAS or direct), not one promised for later.

This script turns a workload into a monthly estimate. Fill in the prices from your own quote.

# Monthly cost estimate for one workload under the OneGov token discount.
# Replace the 0.0 prices with the rates on your own price sheet or quote.
LIST_INPUT_PER_MTOK = 0.0   # <USD per million input tokens>
LIST_OUTPUT_PER_MTOK = 0.0  # <USD per million output tokens>
DISCOUNT = 0.40             # OneGov: 40% off consumption usage

def monthly_cost(requests, in_tokens, out_tokens, growth=1.0):
    """Return (discounted cost, cost at the growth multiple) in USD per month."""
    list_cost = requests * (
        in_tokens / 1e6 * LIST_INPUT_PER_MTOK
        + out_tokens / 1e6 * LIST_OUTPUT_PER_MTOK
    )
    discounted = list_cost * (1 - DISCOUNT)
    return discounted, discounted * growth

# Example workload: 200k requests a month, 3k tokens in, 500 out, 2x growth.
print(monthly_cost(200_000, 3_000, 500, growth=2.0))

The growth argument matters more than the discount. You pay 60% of list, so if usage grows by two-thirds you are back at what the same usage cost at list price. Anthropic says federal civilian usage grew about 20-fold in a year.

Don't write a 40% saving into a multi-year budget until the term is on paper.

For anyone running AI in production for a public agency, the price per token just dropped, but the bill now follows usage, so the plan that survives is the one with a forecast, a cap and a confirmed contract term.