Claude Fable 5 stopped being free today. That's the bill for the tool your ops team built on it last week.

As of July 8, Claude Fable 5 no longer draws from anyone's Pro, Max, Team, or Enterprise subscription limits. Anthropic's top-tier coding model now bills strictly per token through opt-in "usage credits," at $10 per million input tokens and $50 per million output — the most expensive rate on its entire price list. If your firm greenlit an internal tool on Fable 5 in the week since it came back online, this is the morning the free ride ends.

What actually changed on July 7?

Fable 5 relaunched July 1 after an 18-day export-control blackout, and Anthropic included it in Pro, Max, Team, and select Enterprise plans for up to 50% of a user's weekly usage limit — a promotional window meant to let subscribers try the model without a separate bill. That window closed July 7. From July 8 on, Fable 5 requires usage credits enabled under account settings, billed at standard API rates. Anthropic has said the move is operational, not permanent — it's a stopgap while the company builds enough serving capacity to fold Fable 5 back into standard subscriptions. There's no date attached to when that happens.

How much does this actually cost?

Fable 5's per-token rate isn't close to what most construction teams have budgeted for AI tooling. Anthropic's published pricing:

A firm running a document-heavy job — say, feeding a spec book and a stack of RFIs through an agent — pays double the Opus rate and five times the current Sonnet 5 rate for the identical volume of tokens on Fable 5. None of that shows up until the invoice, since the past week's usage was quietly absorbed by the subscription.

Who does this actually hit?

The risk isn't theoretical for firms that moved fast. Fable 5 launched with a genuinely notable capability story — Anthropic's own pre-launch testing had it rewrite a 50-million-line production codebase in a day, the kind of number that gets an ops director's attention for a legacy estimating or ERP migration (we covered that story on July 2). Any tool wired to call Fable 5 by name — a coding agent, a document-migration script, a custom RFI drafter someone stood up over the past week — was running on subsidized usage without anyone checking the meter. Starting today, that same tool either stops working (if credits aren't enabled) or starts generating a real, uncapped bill (if they are).

Should a construction firm downgrade to a cheaper model?

For a one-time, high-value job — migrating a decades-old estimating database, say — Fable 5's premium rate is easy to justify against the cost of doing it by hand, even at $50 per million output tokens. For anything your team runs every day — RFI drafting, submittal review, spec extraction — that math doesn't hold up. Sonnet 5 closes most of the agentic-performance gap to the top tier at a fraction of the cost (we covered Sonnet 5's pricing here), which makes it the better default for routine volume work. Reserve the top-tier model for the rare job that's actually worth the premium.

The takeaway for Monday morning

Before anyone in your shop enables usage credits to keep a tool running, find out which model that tool calls by default. If it's Fable 5 and the task is routine document work, that's a one-line code change to Sonnet 5 that could cut the per-job cost by 80% or more. If it's a genuine one-off — the legacy system nobody wants to touch — the premium rate is the cheap option next to doing it manually. Either way, set a hard usage cap in settings before you fund the credits, not after the first invoice arrives.

Construction AI Brief publishes fresh analysis on what AI news actually means for your jobsite, three times a week. Subscribe at constructionaibrief.com.