Fable 5 Leaves Your Claude Plan on June 22. Here's How to Plan for It

When Anthropic launched Claude Fable 5 on June 9, the headline was benchmarks. The fine print was a clock: Fable 5 is included on Pro, Max, Team, and seat-based Enterprise plans at no extra cost only through June 22. On June 23 it comes off those plans, and continued access runs through usage credits.

Anthropic's stated reason is capacity: "We expect demand for Fable 5 to be very high, and difficult to predict." They say they'll extend the included window if capacity allows, and that the goal is to restore Fable 5 as a standard part of subscriptions "as quickly as we can." No date attached to either promise.

Here's what we actually know, what we don't, and how to plan.

The timeline

• June 9 to 22: Fable 5 included free on Pro, Max, Team, and seat-based Enterprise plans.
• June 23 onward: Fable 5 removed from those plans. Access via usage credits.
• Unchanged: API access ($10/$50 per MTok), consumption-based Enterprise, and cloud platforms (Bedrock, Vertex, Foundry). The June 22 cliff only affects subscription plans.
• Also unchanged: everything else on your plan. Opus 4.8, Sonnet 4.6, and Haiku 4.5 stay exactly where they are.

What we don't know

Credit pricing for subscription users hasn't been detailed beyond "usage credits." The honest answer to "what will my usage cost in credits?" is that nobody outside Anthropic knows yet. Anything you read that puts a number on it is guessing. The API price is the only anchor: $10 per million input tokens, $50 per million output tokens, exactly double Opus 4.8.

What the burn actually looks like

Fable 5 is hungry. Real numbers from the first 48 hours:

• Simon Willison burned $110 worth of tokens in a single day inside a Max subscription, including one session that consumed 78 million tokens (about $99 at API rates).
• One user reported hitting the $200/month Max plan's limits in under 30 minutes.
• Heavy users report individual tasks routinely consuming 500K to 1M tokens.
• Anthropic had to reset rate limits across products on launch day because of demand.

Two reasons it adds up fast. Single requests at high effort can run many minutes and think for a long time before answering, and the model is built for long autonomous sessions, which means multi-hour runs that consume tokens the whole time.

One counterweight: a pre-launch tester reported Fable 5 used roughly half the tokens of Opus 4.8 to complete the same agentic work. The unit price is 2x, but the effective cost on long tasks lands closer to Opus than the sticker price suggests. Short interactive work doesn't get that efficiency win, so chat-style usage really does cost about double.

What to do, by user type

You're on Pro or Max and curious: evaluate now. You have until June 22 to find out whether Fable 5 actually beats Opus 4.8 on your work, for free. Test it on your hardest problems. Anthropic's guidance is explicit that easy tasks undersell it, and early Reddit consensus says output on routine work feels identical to Opus 4.8. If your work is routine, that's your answer: stay on Opus 4.8 after the 22nd and spend nothing extra.

You're on Max and Fable 5 is already your daily driver: frontload. Any big jobs you've been putting off (large migrations, codebase-wide refactors, deep research runs) should happen in the free window. After June 22, queue heavy Fable jobs deliberately and run day-to-day work on Opus 4.8. The "plan with one model, execute with the other" split is emerging as the standard pattern, and it works in both directions.

You manage a team plan: audit who actually needs Fable-class capability before the 23rd. The usage data from this two-week window is the best procurement signal you'll get. If two engineers run week-long agentic jobs and eight people use Claude for everyday work, you're looking at credits or API access for two people, not a plan change for ten.

You're API-only: nothing changes for you on June 22. Your decision is the standard one: does Fable 5 justify 2x Opus pricing on your workload? Run both on identical tasks and measure cost per completed task, not cost per token.

Three ways to control cost when the meter starts

1. Use effort levels. Fable 5 has five (low through max) with a wide cost spread between them. Anthropic recommends starting at high, not xhigh. Independent testing found the same task varied from about 10 cents to about 72 cents depending on effort level.
2. Match the model to the task. Fable 5's lead over Opus 4.8 grows with task length and complexity. Inverting that: the shorter and more routine the task, the less you get for the 2x price. Default to Opus 4.8 or Sonnet 4.6; escalate to Fable 5 when a task is genuinely hard.
3. Watch the window, not the docs. Anthropic said they'll extend the free period "if capacity allows." Given they reset rate limits on day one from demand strain, don't plan around an extension. Treat June 22 as real.

The bigger picture

It's fair to read this launch as a preview of how frontier-model access will work from now on: the newest tier costs real money, and subscriptions cover the tier one notch down. The top Hacker News thread on the launch called the two-week window the "free sample method," and the comparison isn't wrong.

But the practical version of that story is less dramatic. Opus 4.8 is still on your plan, still excellent, and still what Fable 5 falls back to when its own safety classifiers fire. For most work, the model you already have is the right tool. The two weeks of free Fable 5 are best spent figuring out exactly which slice of your work is the exception.

You have until June 22. Test accordingly.

FAQ

Is Claude Fable 5 free?

Temporarily, yes - on paid plans. Per Anthropic's launch announcement (accessed June 11, 2026), Fable 5 is included on Pro, Max, Team, and seat-based Enterprise plans at no extra cost through June 22, 2026; on June 23 it comes off those plans and continued use requires usage credits. There is no free-tier access, and API usage is billed from day one at $10 per million input tokens and $50 per million output tokens.

Sources: Anthropic's launch announcement, pricing docs, Simon Willison's first-day writeup, and the Hacker News launch thread.