Anthropic gives government 40% off Claude. The $1 era is over — the metered bill is here.

Anthropic and the U.S. General Services Administration expanded their OneGov agreement on October 27, 2026: government buyers now get a 40% discount on Claude token usage, with no minimum order, across Claude Enterprise, Claude for Government, and the Claude API. Eligibility extends beyond federal agencies to state, local, tribal, and territorial governments for the first time. [GSA news release, Oct 27, 2026; Anthropic announcement via Unite.AI, Oct 27, 2026]

The timing is the story. The $1 OneGov offer under the 2025 agreement ended October 31, 2026. Since November 1, government buyers are on a metered bill at 40% off list. That is still a steep discount. It is not $1. [GSA, Oct 27, 2026; GovConFeed, Oct 27, 2026]

Here is the number that matters: 40% off every token, no minimum spend, pay only for what you use. The discount applies to direct purchases now; purchases through cloud providers get the same discount "in the coming months," so agencies buying through AWS or Google Cloud today are still waiting on their channel. [Anthropic via Unite.AI, Oct 27, 2026; BigGo Finance, Oct 27, 2026]

The per-unit math

Apply 40% off to Anthropic's published October 2026 list rates:

[Anthropic list pricing, October 2026; 40% discount per GSA, Oct 27, 2026]

Cached input scales the same way — Sonnet 5.5's $0.10 cache-read rate becomes $0.06, Opus 5.5's $0.20 becomes $0.12, Haiku 5.5's $0.01 becomes $0.006. For agent workloads where cache reads dominate the bill, the discount lands hardest exactly where the tokens pile up.

Now put a real workload through it — 10 million input tokens and 5 million output tokens, standard rates, no cache, no batch:

Scale that to an agency running 100 million input and 50 million output tokens a month on Sonnet 5.5: $700 list becomes $420 — $280 a month, $3,360 a year, for filing one procurement form. Anthropic estimates the agreement will save the government about $1 billion in total; GSA says its OneGov strategy has already generated more than $1.8 billion in savings, $1.6 billion of it through "super intelligence" agreements. Treat both figures as the vendors' case, not an audit — neither has published the underlying usage math. [Anthropic via Unite.AI, Oct 27, 2026; GSA, Oct 27, 2026]

flowchart LR
    A["List price<br/>Sonnet 5.5: $2 / $10"] --> B["OneGov discount<br/>−40%, no minimum"]
    B --> C["Government price<br/>Sonnet 5.5: $1.20 / $6"]
    C --> D["Opus 5.5: $2.40 / $12<br/>Haiku 5.5: $0.06 / $0.30"]

What changed, and what did not

Three things are genuinely new:

• Eligibility widens. The old deal was federal. The new deal covers federal (executive, legislative, judicial), state, local, tribal, and territorial governments — the SLTT category that includes state IT departments, county offices, and tribal governments. Roughly 500 U.S. federal, state, and local organizations already use Claude, per Anthropic. [GSA, Oct 27, 2026; Pandromeda, Oct 27, 2026]
• Every product surface, first time. The discount covers Claude Enterprise, Claude for Government, and the Claude API together — Anthropic calls this a first. Previously, government pricing was fragmented by product. [Anthropic via Unite.AI, Oct 27, 2026]
• Consumption-based, no floor. No minimum orders, no upfront commitment. Agencies pay for usage consumed, through direct purchase, resellers, or — soon — major cloud platforms. [GSA, Oct 27, 2026]

And one thing that did not change: this is a government-channel price, not a list-price cut. If you are not buying through GSA's Multiple Award Schedule at the OneGov tier, your Claude bill is untouched. Sonnet 5.5 is still $2/$10 for everyone else. Do not let a headline about a 40% cut rewrite your private-sector budget.

The context the headline skips

This is the second government pricing reset in a week of list-price warfare. OpenAI's GPT-6.1 Terra launched October 26 at $1/$5 per million tokens — half of Sonnet 5.5's list price — and reporting on October 25–27 describes OpenAI offering large enterprise accounts steep private discounts, with one U.S. bank paying under $7 per ChatGPT seat against a $20 list rate. The pattern across both labs is now unmistakable: the published price is a ceiling, and the real price is negotiated by segment. Government buyers get 40% off by procurement channel; enterprise buyers get private deals; everyone else pays list. [The Information via Pivot News, Oct 25–27, 2026]

Anthropic's own usage claims point the same way: weekly Claude for Government active users up about eightfold since February, federal civilian usage up about 20-fold in a year — vendor figures, not independently verified, but consistent with a market where the government channel is now a real revenue line, not a pilot. [Anthropic via BigGo Finance, Oct 27, 2026]

What to do about it

flowchart TD
    Q{"Do you buy through a government channel?"}
    Q -->|Yes, direct from Anthropic| D["Claim the OneGov rate now: 40% off token usage, no minimum, via GSA MAS"]
    Q -->|Yes, via a cloud provider| W["Wait for the cloud-channel rollout in the coming months — do not re-budget yet"]
    Q -->|No| N["Nothing changes: Sonnet 5.5 stays $2/$10 list for you"]
    D --> M["Meter your usage: the $1 flat era ended Oct 31 — forecast on tokens, not seats"]
    W --> M

If you're a heavy user in government — an agency already running Claude at volume: the move from $1 flat to metered-at-40%-off means your November bill is now a function of tokens, not headcount. Forecast it now: take last month's token volume, apply the OneGov rates above, and compare against whatever the $1 arrangement actually cost you. If your token volume is small, the metered bill may still be trivially cheap; if it is large, the 40% discount is doing real work against list — Sonnet 5.5 at $1.20/$6 undercuts OpenAI's Terra only on input, not output, so route output-heavy work with care.

If you're a light user in government — a small office, a county team, occasional use: the no-minimum term is the whole deal. You can now buy Claude at 40% off with no commitment, which the old federal-only arrangement never offered you at all. SLTT eligibility is the news, not the percentage.

If you run a team outside government: do not re-price anything — but do read the signal. When a vendor offers one buyer segment 40% off list and another segment private seat discounts, list price has stopped being the price. If your Claude or ChatGPT bill is material, the question to take into your next renewal is not "will list prices fall?" but "what segment discount do we qualify for, and what do we have to give up to get it?" That is a procurement conversation, and it is worth having before your December invoices.

Every figure dated, every claim sourced. The AI Bill is an independent practice — no vendor sponsorship, no affiliate links, no paid placement.

Sources: GSA news release, "GSA, Anthropic Reach OneGov Agreement on Discounted, Consumption-Based Pricing for Claude" (Oct 27, 2026 — 40% discount, no minimum orders, eligibility across federal/state/local/tribal, $1.8B OneGov savings claim); Anthropic announcement via Unite.AI (Oct 27, 2026 — discount across Claude Enterprise, Claude for Government, and Claude API; cloud-provider rollout in coming months; ~$1B savings estimate; ~500 government organizations); Pandromeda (Oct 27, 2026 — FedRAMP High authorization, DoD IL4/IL5 on AWS GovCloud); BigGo Finance (Oct 27, 2026 — usage growth claims, dedicated agency deployment program); GovConFeed (Oct 27, 2026 — $1 offer expiry Oct 31); Anthropic list pricing, October 2026 (Haiku 5.5 $0.10/$0.50, Sonnet 5.5 $2/$10, Opus 5.5 $4/$20, Fable 5.1 $10/$50); The Information via Pivot News (Oct 25–27, 2026 — OpenAI enterprise discount reporting, context only).

Next: Book a free 45-minute teardown — screen-share of your last 3 months of AI spend, fixed-fee quote before anything starts.