The US General Services Administration (GSA), the agency that buys technology for the federal government, announced on 27 October 2026 a new OneGov agreement with Anthropic: a 40% discount on token-based use of Claude models, with no minimum purchase, for federal, state, local and tribal governments. It replaces a flat offer of $1 per user that expires on 31 October 2026. Australian businesses cannot use the discount, but the shift from per-seat licences to pay-per-use pricing is worth understanding before your own AI contracts come up for renewal.
What happened
GSA published its release, "GSA, Anthropic Reach OneGov Agreement on Discounted, Consumption-Based Pricing for Claude", on 27 October 2026. It says government buyers get 40% off token-based usage across the Claude models available to the government, including FedRAMP-authorised environments for sensitive unclassified work. FedRAMP is the US government's security authorisation program for cloud services.
Consumption-based pricing means the agency pays for what the model processes, measured in tokens (small chunks of text, roughly a word or part of a word), instead of a fixed fee per person. GSA says there are no minimum orders: agencies pay only for what they use.
Anthropic, the company that makes Claude, posted its own announcement the same day. It says the 40% discount covers Claude Enterprise, Claude for Government and the Claude API (the connection software developers use to build Claude into their own systems), and that it also extends to territorial governments. For the first time, Anthropic says, the discount applies across every product surface: when agencies buy directly today, and through cloud providers in the coming months.
The older deal worked differently. GSA's OneGov page lists the earlier Claude offer at $1 per user for federal agencies only, expiring on 31 October 2026. The new offer runs to 31 January 2029.
Key details
| Point | Earlier offer | New agreement |
|---|---|---|
| Pricing model | $1 per user | 40% off token-based usage |
| Who can buy | Federal agencies only | Federal, state, local and tribal entities (Anthropic adds territorial governments) |
| Products | Claude for Enterprise and Government | All Claude models available to government, including Claude Enterprise, Claude for Government and the Claude API |
| Commitment | Per-user licences | $0 commitment and $0 minimum orders |
| How to buy | GSA Multiple Award Schedule | Anthropic direct, resellers and supported cloud marketplaces, through the GSA Multiple Award Schedule |
| Expiry | 31 October 2026 | 31 January 2029 |
Anthropic gives some numbers of its own. It says approximately 500 US federal, state and local government organisations now use Claude, that weekly active users of Claude for Government are up about eightfold since February, and that use across federal civilian agencies has grown about 20-fold in a year. Anthropic estimates the discounts will save the government $1 billion. GSA does not repeat that estimate; it says its OneGov strategy has produced more than $1.8 billion in savings since it began, about $1.6 billion of it through its super intelligence agreements.
On security, Anthropic says Claude for Government runs at a FedRAMP High authorisation level, that Claude on Amazon Bedrock and on Google Cloud's Vertex AI is authorised for FedRAMP High, and that Claude Enterprise is in process for FedRAMP Moderate through the FedRAMP 20x program.
Why it matters
This is a procurement decision, not a new law, but procurement shapes markets. When the US government's central buying agency moves from a per-seat licence to metered use with no minimum, it signals how AI suppliers expect to sell to large organisations: the bill follows the work the model does, not the number of people who have a login.
It also fits a pattern we have followed on this blog. Model prices keep moving, sometimes down sharply, as with Anthropic's price cut on its small Haiku model, and sometimes in new shapes, as with DeepSeek's peak and off-peak pricing. A business that signed a fixed AI contract a year ago may now be paying for a structure the market has moved past.
The wording is worth noticing too. GSA's release uses the term "super intelligence" where most organisations write AI. The tools it describes are the same Claude models and the same API that businesses use, so the label changes nothing about what is being bought.
Finally, the government's own take-up says something about trust. Anthropic says about 500 government organisations already use Claude, and the agreement covers environments authorised for sensitive unclassified work. For a supplier your business may depend on, that is one more data point alongside what its share-listing filing showed about the company's risks and growth.
What this means for businesses
The discount is for US government buyers only. An accounting practice in Brisbane or a wholesaler in Auckland pays Anthropic's normal commercial prices, and neither announcement mentions any change to those. What you can take from it is the pricing question the US government just answered for itself: are you paying for seats or for use, and which one suits how your team works?
A per-user plan is simple to budget, but you pay for every licence, including the staff who open the tool once a month. Token pricing tracks actual use, which suits uneven or automated work (summarising inbound emails, drafting quotes, sorting support tickets), but the bill can grow quietly if an integration or an AI agent runs more often than anyone expected. Our guide to permissions, data access and cost for AI agents covers how to put limits on that.
A short checklist for the next time an AI subscription or contract renews:
- List every AI tool the business pays for, who uses it and how often. Unused seats are the first saving.
- Ask each supplier whether usage-based pricing is available, and what the minimum commitment is.
- If you use an API, set a monthly spending limit and an alert in the supplier's console before you scale up.
- Check where your data is processed and stored, and whether the plan's security terms suit the information you put into it.
- Read the renewal terms: how much notice the supplier gives before changing prices, limits or models.
- Keep the integration simple to switch, so a better price or model elsewhere does not mean a rebuild.
Someone has to own this list. In many small businesses it falls to whoever set up the first account; if nobody has the time, a recruitment agency such as Talynx can help you find IT staff who can manage software licences and suppliers.
If the answers point to a change, the work is usually in the plumbing: moving a team from seats to API access means connecting the model to your CRM, inbox or job system, with logging and cost controls. Our business systems and integrations work covers that, and IT consulting can help you compare plans and contracts before you commit. You can book a free first consultation and get a written quote before any work starts.
Frequently asked questions
Can an Australian business get the 40% Claude discount?
No. The OneGov agreement is for US government buyers: federal, state, local and tribal entities in GSA's listing, with territorial governments added in Anthropic's post. Commercial customers, in Australia or anywhere else, buy Claude on Anthropic's normal plans and prices.
Does this change the price my business pays for Claude?
Neither announcement mentions any change to commercial plans or prices. What it does show is the pricing model the US government chose for itself: pay per token, no commitment, no minimum order. That is a useful benchmark when you review your own AI subscriptions and contracts.
What is the difference between per-user and per-token pricing?
Per-user pricing charges a fixed fee for each person with a licence, whether they use the tool or not. Per-token pricing charges for the text the model reads and writes, so the bill follows actual use. Light, uneven use usually favours tokens; heavy daily use by the whole team can favour seats.
Why does the GSA release say super intelligence instead of AI?
The GSA release of 27 October 2026 uses the term super intelligence where most organisations would write AI, including in its subhead about advanced super intelligence tools. The products it describes are the Claude models and the Claude API, the same tools businesses use.
How long does the new US government agreement run?
GSA's OneGov page lists the new Claude offer with an expiry of 31 January 2029. The earlier offer of $1 per user for federal agencies expires on 31 October 2026, and GSA says it renews agreements where appropriate as offers expire.
Key takeaways
- GSA and Anthropic announced on 27 October 2026 a 40% discount on token-based Claude use for US federal, state, local and tribal governments.
- It replaces a $1-per-user federal offer that expires on 31 October 2026; the new offer is listed to 31 January 2029.
- There is no commitment and no minimum order: buyers pay for what they use, directly, through resellers or through cloud marketplaces.
- Australian and other commercial customers are not covered, and neither announcement changes commercial prices.
- The useful lesson for a small business is the pricing model: check whether you pay for seats or for use, and which suits your team.