Claude Fable 5: What It Is, Why It Vanished, and What Its Return Means

Claude Fable 5 is Anthropic's first generally available Mythos-class model. In Anthropic's published positioning, Mythos-class sits a full tier above the Opus-class models most people already know, so Fable 5 is effectively the most capable Claude that everyday users and developers can reach without special access. It was introduced on June 9, 2026 alongside its bigger sibling, Claude Mythos 5.

The headline strength is long-horizon agentic work: tasks that require a model to plan, use tools, write and revise code across many steps, and stay coherent over a long session without losing the thread. This is exactly where Fable 5 opens its widest lead over Opus 4.8. If your work looks like “here is a big, messy engineering problem, go solve it end to end,” Fable 5 is built for that.

Fable 5 is available through Claude.ai, the Claude Platform API (model id claude-fable-5), Claude Code, Claude Cowork, and Amazon Bedrock on AWS. It is a premium, paid model, not part of the free tier.

Few model launches have had a stranger first month. Here is what happened, in order:

• June 9, 2026, launch. Anthropic introduces Claude Fable 5 and Claude Mythos 5, its new Mythos-class frontier models.
• June 12, 2026, offline. Just three days later, both models are forced offline under U.S. export-control restrictions, after administration officials say they pose severe cybersecurity risks. Access is cut across Anthropic's surfaces.
• Mid-to-late June, the gap. Users hit “Fable 5 currently unavailable” and “invalid or inaccessible model” errors. Search interest in “fable 5 coming back,” “fable 5 shutdown,” and “claude fable 5 ban” spikes into breakout territory.
• July 1, 2026, restored. The U.S. Commerce Department lifts the restrictions. Commerce Secretary Howard Lutnick informs Anthropic of the decision, and Fable 5 and Mythos 5 return across Claude.ai, the Claude Platform, Claude Code, and Claude Cowork.

So to answer the most-searched question directly: yes, Fable 5 is back. It was never a bug or a quiet retirement. It was a government policy action that has since been reversed.

The pause was about export controls and national-security concerns, not model quality. Senior U.S. administration officials judged that Mythos-class models like Fable 5 and Mythos 5 were capable enough to pose severe cybersecurity risks if misused, and the restrictions required Anthropic to cut off access while the policy was in force.

This is why the “jailbreak” searches surged. The worry driving the action was that a sufficiently powerful frontier model, if its safety guardrails were bypassed, could meaningfully help with dangerous cyber or scientific tasks. The restriction was a precautionary measure at the policy level rather than evidence that Fable 5 had actually been jailbroken in the wild.

To get the models restored, Anthropic agreed to keep collaborating with the government on protocols, standards, and future releases for Mythos, Fable, and later models, and to inform the U.S. government of any malicious activity. Those commitments, plus the safeguards built into Fable 5 itself, formed the basis for lifting the ban.

On Anthropic's published, unsafeguarded benchmarks, Fable 5 leads Opus 4.8 on every measure both models report, and the gap is widest on hard, long-horizon coding.

The pattern is clear: on everyday tasks the two models are close, but as problems get longer and more agentic, Fable 5 pulls away. The roughly 11-point gap on SWE-bench Pro and the near-doubling on FrontierCode Diamond are the numbers that justify Fable 5's premium for engineering teams.

One caveat worth understanding: these are the unsafeguarded scores. In real use, Fable 5's safety routing can hand certain flagged requests to Opus 4.8, so a small share of sensitive sessions effectively run on the Opus model instead.

Fable 5 costs about twice as much per token as Opus 4.8. For heavy automated workloads that price difference adds up fast, which is why many teams reserve Fable 5 for the tasks that actually need it and run everything else on Opus 4.8. See our Claude pricing guide for the full model lineup and costs.

The 30-day data-retention requirement is the other decisive factor. Organizations that need zero data retention for compliance cannot meet that requirement on Fable 5 today, so Opus 4.8 remains the model for those workloads.

Fable 5 and Mythos 5 are both Mythos-class models released on the same day, and both were pulled and restored together. The simplest way to think about them: Mythos 5 is the flagship of the class, and Fable 5 is the Mythos-class model made generally available for broad, everyday use. If you are choosing a model to actually build with today, Fable 5 is the one you will reach for, with Opus 4.8 as the cheaper, zero-data-retention alternative below it.

Because both carry the same class of capability, they attracted the same regulatory scrutiny, which is why the export-control action named both and why their availability has moved in lockstep.

The safety design is central to why Fable 5 was allowed back. Rather than answer every request directly, Fable 5 routes flagged high-risk requests to Claude Opus 4.8. Specifically, prompts that look like they involve cybersecurity exploitation, biology, chemistry, or model distillation can be handled by the more conservative Opus model instead of Fable 5. Anthropic says this is designed to affect only a small minority of sessions, and that these Opus-fallback responses are not billed at Fable prices.

This is the practical answer to the flood of “fable 5 jailbreak” searches. The concern that drove the ban was misuse of a very capable model, and the built-in routing plus Anthropic's monitoring commitments are how that concern is being managed. For normal users doing normal work, Fable 5 behaves like any other Claude model, just with extra guardrails on a narrow set of sensitive topics.

Now that Fable 5 is restored, here is where to find it:

• Claude.ai: open the model picker on a paid plan and select Fable 5.
• Claude Platform API: call the claude-fable-5 model id.
• Claude Code: update to the latest version, then choose Fable 5 as your model for heavy coding sessions.
• Claude Cowork and AWS Bedrock: Fable 5 is available on both, with Bedrock offering it with built-in safeguards.

If you still see “Fable 5 currently unavailable” or “invalid or inaccessible model claude-fable-5,” the fix is almost always to update your client and refresh the model list. Older builds cached the model catalog from the pause, when Fable 5 genuinely was not selectable. For a full walkthrough of Claude's tools, see our guides on Claude prompts and using Claude for coding.

The honest answer for most people is: default to Opus 4.8, and escalate to Fable 5 for the hardest jobs. Opus 4.8 is half the price, supports zero data retention, and is a genuinely excellent model for writing, analysis, and everyday coding. You will not notice Fable 5's advantage on a quick email or a simple script.

Where Fable 5 earns its premium is the long tail of difficult work: multi-step agentic tasks, large refactors, and complex engineering problems where its benchmark lead means fewer wrong turns and less human correction. On those jobs the higher token price is often cheaper overall, because the model finishes the task with less back-and-forth. If you are comparing options across providers, our ChatGPT vs Claude comparison puts these models in context.

The bigger takeaway from the Fable 5 saga is about the landscape itself. Frontier models are now capable enough that governments are actively involved in when and how they ship. Expect more of this: powerful launches, occasional pauses, and fast-moving news. The practical move is to build on the model that fits each task today, and to stay informed as availability changes.