Anthropic Ended Its Free Window for a Model It Cannot Legally Serve
On June 23, Anthropic began charging $50 per million output tokens for Claude Fable 5. The model has been offline since June 12 under US export-control suspension. The 14-day free window was marketed at launch; the model was available for the first 3 of those days. The refund window for subscribers who upgraded closed June 20, three days before the billing started. Anthropic is now collecting non-expiring credits for a product under open-ended government suspension — and setting a structural pricing precedent that may outlast the suspension itself.
A timeline.
June 9: Anthropic launches Claude Fable 5 and Mythos 5. Subscribers on Pro, Max, Team, and Enterprise plans get 14 days of complimentary access before the model moves to separate per-token billing at $10/M input, $50/M output.
June 12: The Bureau of Industry and Security serves Anthropic a directive under ECRA § 4817(b)(1) and EAR § 744.22(b). Anthropic cannot verify customer nationality in real time across AWS Bedrock, Google Cloud, Microsoft Foundry, and Snowflake. Both models go offline for all users globally. The export-control suspension begins on day 3 of the 14-day free window.
June 20: The refund window closes. Subscribers who upgraded their plan between June 9 and June 14 specifically to access Fable 5 are eligible for prorated refunds — if they apply by today. The model has been offline for 8 days. Many do not apply in time.
June 23: Anthropic executes the pricing transition as planned. Fable 5 is still offline. Subscribers who want continued access when the model returns must purchase usage credits at $50/M output — twice the cost of Claude Opus 4.8, and $20/M more expensive than GPT-5.5 per output token. On June 23, 2026, Anthropic is collecting payment for a product under active US government suspension with no disclosed restoration date.
Today, day 13: The model remains unavailable to every user in the world.
The Hacker News thread on the June 23 transition surfaced the obvious objection: "You're now collecting credits for a model you can't legally serve." Anthropic's answer, implicit in its policy, is that the credits are non-expiring — deferred delivery, not cancelled delivery. That is technically true and practically unsatisfying. "Non-expiring" is not a delivery commitment. It does not specify a timeline, a restoration condition, or what Anthropic owes subscribers if Fable 5 does not return globally.
The official framing from the White House is that the path forward is straightforward — Anthropic needs to fix the underlying vulnerability that triggered the BIS directive, and the export restriction follows. Eleven days of silence from BIS suggests either the fix is harder than it sounds, or the government's definition of "fixed" involves something other than a technical patch.
Two structural restoration milestones are on the calendar. July 8: Anthropic's updated privacy policy takes effect, which includes government-issued ID and biometric collection — read by most observers as groundwork for a US-citizen-verified restoration that satisfies BIS without requiring the directive to be lifted. August 1: The 60-day deadline for the Trump AI executive order's frontier model framework, which may establish formal criteria for when export controls apply to AI APIs and how compliance is demonstrated.
Polymarket was pricing Fable 5 restoration at roughly 55-57% before July 1. Those are not the odds of a model that is two days from returning.
The effective free trial was three days
This fact has not been centered in coverage. Anthropic announced a 14-day complimentary window. Subscribers who signed up on June 9 got three days of access before the model disappeared. If you signed up on June 10, you got two. If you signed up on June 12 after 5:21 PM ET, you got approximately zero.
The 14-day window was a marketing commitment that the US government rendered impossible to fulfill on day 3. Anthropic is not responsible for BIS serving a surprise directive. It is responsible for how it has handled the period since — and the decision to close the refund window on June 20, proceed with the billing transition on June 23, and collect non-expiring credits for an unavailable product represents a series of choices that favored its cash position over its subscriber experience.
The eligible refund window — new upgrades between June 9 and June 14 only, application deadline June 20 — was designed before anyone knew the model would be suspended on day 3. Its narrow eligibility and eight-day application window were reasonable in a world where the model was running. In the world that actually happened, they look like a deadline set by someone who did not update for events.
The structural story that will outlast the suspension
Here is what matters after Fable 5 returns, whenever that is.
Anthropic is the first major frontier AI lab to implement explicit pay-above-subscription pricing for its newest model. OpenAI has used usage caps and rate limits within subscriptions, but has not created a separate billing tier above the subscription level. Google has not either.
If this model holds — if subscribers accept that the newest Anthropic frontier model is not included in their $20/month or $100/month plan and requires per-token credits at $50/M output — every subsequent Anthropic release will be priced the same way. The Pro subscription becomes a floor, not a ceiling. You pay for access to the platform; you pay separately for access to the best model on the platform.
The $50/M output price is also a market signal. Fable 5 at $50/M is 67% more expensive than GPT-5.5 at $30/M on a per-output-token basis. Anthropic is asserting that Fable 5 is worth that premium. Before June 12, that assertion was testable — users could evaluate the model and decide. Since June 12, the assertion is entirely theoretical. The market is being asked to buy credits for a capability it has not been able to access for two weeks.
I think the pricing survives if the model returns soon and performs. If Fable 5 comes back, competes with GPT-5.5 on real tasks, and enterprise customers find the 67% output premium justified by capability, the credits model will be validated. The structural precedent will stand.
If Fable 5 returns geographically restricted — US-only access under the July 8 ID verification framework — the $50/M output price is for a product that most of the world cannot use. The pricing and the restriction, together, would be the narrowest definition of "frontier model access" any major lab has ever charged for.
And if Fable 5 does not return before the credits model creates a consumer protection inquiry, the structure that was supposed to establish a new pricing tier will instead become a case study in how not to price a model the government took offline on day 3 of its free trial.