Fable 5 is back in Claude after 19 days offline

Claude Fable 5 is available again. The model was shut off worldwide on 12 June 2026 after a U.S. export control order, and was redeployed on 1 July 2026 — after 19 days offline. Today you can reach it in Claude.ai, Claude Code, Claude Cowork and through the API.

This is the story of what happened, what is different now, and exactly how to get access.

How to get access to Fable 5 today

Since 20 July 2026 access has been split permanently, according to Anthropic's help centre:

• Max and Team Premium: Fable 5 is included in the plan, up to 50% of your weekly usage limits.
• Pro and Team Standard: access runs on usage credits (pay as you go). Existing users received a one-time $100 credit when the switch was made.
• API (Claude Platform): $10 per million input tokens and $50 per million output tokens. Prompt caching cuts input cost by up to 90%. Choosing US-only inference adds a 1.1x multiplier.
• Cloud platforms: the model is also available through AWS, Google Cloud and Microsoft Foundry.

In practice: on Max you simply pick Fable 5 in the model selector. On Pro you have to enable usage credits first — otherwise the model won't show up as an option at all.

One detail worth knowing: biology and cybersecurity queries are automatically routed to the Opus models for safety reasons. You are not charged Fable 5 rates for those rerouted calls.

What happened: 19 days offline

9 June — the launch

On 9 June 2026, Anthropic launched Claude Fable 5, the company's new flagship in the so-called Mythos class. The numbers were staggering. Stripe reported that Fable 5 compressed months of work into days, migrating a 50-million-line codebase in a single day — something an entire team would otherwise have needed two months for.

Alongside Fable 5 came Mythos 5: the same underlying model, but with certain safety guardrails lifted. It was made available only to cybersecurity professionals and critical infrastructure providers through Project Glasswing.

12 June — the directive

Late on Friday 12 June, at 5:21 PM ET, Anthropic received a directive from the U.S. government citing national security. The content: block access for all foreign nationals — both inside and outside the United States, including Anthropic's own employees.

Because the company cannot separate foreign users from the rest in real time, the result was a worldwide shutoff of both models. Behind it was a report from Amazon's security researchers describing a way to bypass — "jailbreak" — Fable 5 so it would analyse code and surface software vulnerabilities.

26–30 June — the restrictions lift

On 26 June the government approved restoring Mythos 5 for a set of U.S. organisations. Four days later, on 30 June, export controls on both Fable 5 and Mythos 5 were lifted entirely.

1 July — the redeployment

The next day Fable 5 was back for users worldwide. Anthropic had by then added a new safety classifier that blocks the reported technique in over 99% of cases and reroutes those requests to Opus 4.8 instead.

Mythos 5 is still handled separately: it is available to a selection of U.S. organisations, and Anthropic continues to work with the government on widening the Glasswing programme.

What Fable 5 can actually do

This was no marginal update. Fable 5 took top scores on several heavyweight benchmarks — Cognition's FrontierCode, Hebbia's Finance Benchmark and multiple tests in physics and analytics.

• Software engineering: handles entire codebases autonomously, and holds together tasks that run across several days
• Vision: extracts precise numbers from scientific figures and rebuilds web apps directly from screenshots
• Research: Mythos 5 accelerated protein design roughly 10x, and scientists preferred the model's molecular biology hypotheses in 80% of cases in blind tests

If you want to compare it with the rest of Anthropic's line-up — Opus 5, Sonnet 5 and Haiku 4.5 — we have collected pricing, context windows and benchmarks in our guide to Claude.

What it means for businesses and developers

The model coming back doesn't make the lesson any smaller. Nineteen days is a long time if your product rests on a single provider, and the shutdown is believed to be the first time a leading AI company has taken a publicly deployed model offline following direct intervention from the federal government.

A few things worth taking away:

• Don't lock yourself into a single model. An abstraction layer that lets you swap providers costs little now and saves you on the day access changes.
• Have a plan B. Which parts of your business grind to a halt if your primary model disappears for three weeks?
• Cost the work, not just the capability. Fable 5 is the most expensive model in the family. For many workflows Opus 5 or Sonnet 5 is plenty — at a fraction of the price.
• Follow the rules, not just the releases. What a model is allowed to do is increasingly governed by law and export controls — especially for those of us in Europe.

From tech product to security matter — and back

It says something about where we are. AI went from tech product to national security matter in three days, and back again in nineteen. The question is no longer just what the models can do, but who is allowed to use them — and how fast that can change.

For Swedish and European companies it underlines something we notice more and more in our consulting work: those who understand both the technology and the rules around it will stand considerably steadier in the years ahead.