All you need to know about Claude Fable 5

On June 9th, 2026, Anthropic announced its most powerful models to date: Claude Fable 5 and Claude Mythos 5. With Fable 5, Mythos‑class capability is suddenly available to any team building on the Claude stack.

The launch also marks a shift in how Anthropic is presenting its top‑end systems. Fable 5 is the company’s latest frontier model that teams are expected to run in real workloads, but it arrives with clearer controls over how it behaves and how it is used than earlier Claude generations.

When will Fable 5 be back online?

As of June 16, 2026, Anthropic has not given a firm public date for when Claude Fable 5 will come back online. However, Anthropic is reportedly in discussion with White House officials to coordinate a solution.

On June 12, Anthropic suspended access to both Claude Fable 5 and Claude Mythos 5 after receiving a US export‑control directive that required the company to shut down the models globally rather than risk violating restrictions on foreign users.

Anthropic has said it disagrees with regulators regarding the underlying risk and that it is working with the US government to restore access, but so far it has only described the suspension as temporary and has not committed to a specific restoration timeline.

What are Claude Mythos and Claude Fable?

Claude Mythos 5 is Anthropic’s latest model in its new Mythos‑class generation, a tier that outperforms earlier Opus‑class systems on coding, knowledge work, and security‑sensitive analysis. It’s powerful enough that Anthropic limits the less‑restricted Mythos configuration to a small group of vetted security, financial, and critical‑infrastructure customers rather than the general public.

Claude Fable 5 is the generally available configuration of that same Mythos‑class model. While it shares training and weights with Claude Mythos 5, it adds a dedicated safety layer that blocks or downgrades responses in high‑risk domains like cybersecurity and biology, and routes certain prompts to Claude Opus 4.8, a weaker model, instead. In practice, that makes Fable 5 Anthropic’s most capable model that enterprises can adopt as a default “frontier” option for demanding reasoning, software engineering, and long‑horizon agentic tasks.

Operationally, Fable 5 sits at the top of the Claude lineup—above Haiku, Sonnet, and Opus—and is exposed through the usual channels: the Claude API, Anthropic’s UI, and major cloud and data platforms. That lets teams plug Mythos‑class capabilities into existing stacks on AWS, Databricks, Snowflake, and similar environments while relying on Anthropic’s safety layer and platform‑level governance to handle access control, logging, and cost management.

A quick history of Mythos

Before Mythos 5, Anthropic had already spent several Claude generations pushing into higher‑risk territory: helping teams find software vulnerabilities, reason about complex systems, and work with sensitive cyber and bio content under tight constraints. As those capabilities improved, Anthropic began carving out a separate “Mythos‑class” tier for models that were powerful enough to require more limited access and heavier governance, rather than simply treating them as another Opus‑style upgrade.

Claude Mythos 5 started off with an unplanned public frenzy: Before its official April announcement, the details around Anthropic’s new model were accidentally leaked to the public in a cache of unsecured documents. 

The cache included more than 3,000 assets, including a draft blog post that referred to Mythos as “by far the most powerful AI model we’ve ever developed.” Anthropic later admitted that they were, indeed, working on a new and improved model, and that, “given the strength of its capabilities, [they’re] being deliberate about how [they] release it. “

Anthropic first released the Claude Mythos Preview on April 7th, 2026, as part of Project Glasswing, an initiative aimed at helping critical software and infrastructure providers harden their systems using Mythos‑class capabilities. Rather than a broad public launch, the company limited access to about 50 security and infrastructure partners that included Apple, Google, and NVIDIA, among others. On June 2nd, they expanded the program to roughly 150 organizations under closely managed terms.

Mythos would remain the less‑constrained configuration for trusted users, while Claude Fable 5, the safer, generally available version, was developed in parallel. 

Capabilities and benchmarks

Anthropic positions Fable 5 as state of the art on nearly all of its tested benchmarks, with the performance gap widening as tasks become longer and more complex.

Software engineering

Fable 5 is designed for long-horizon engineering work rather than short, chat-style assistance. According to Anthropic, Stripe used the model to complete a codebase-wide migration of a 50-million-line Ruby codebase in one day—work that would otherwise have taken a team more than two months. Anthropic also says that Fable 5 leads frontier models on Cognition’s FrontierCode evaluation, even at medium effort, while using tokens more efficiently than prior Claude models.

Knowledge work

Fable 5 boasts the highest score of any model on Hebbia’s finance benchmark for senior-level reasoning, with “substantial gains in document-based reasoning, chart and table interpretation, and problem solving.”

IMC Trading, a global proprietary trading firm and market maker, found Fable 5 especially strong on factual lookup, conceptual reasoning, root-cause analysis, and expected-value analysis in trading scenarios. This indicates that beyond being a coding model, Fable 5 is useful for high-complexity analytical work in finance and enterprise research.

Vision and multimodal work

Fable can extract precise values from detailed scientific figures, reconstruct a web app’s source code from screenshots, and handle multimodal tasks with less scaffolding than previous Claude models required.

In the announcement post for Fable 5, Anthropic demonstrated how, with its enhanced vision, it was able to breeze through Pokémon FireRed with a minimal harness, while other models floundered even with additional support. 

Memory and long context

Anthropic says Fable 5 stays focused across millions of tokens and can improve outputs using its own notes in long-running tasks. In the company’s internal testing on the game Slay the Spire, persistent file-based memory improved Fable 5’s performance three times more than it improved Opus 4.8, and Fable reached the game’s final act three times more often.

Data retention, governance, and risk

Claude Fable 5 is not just a more capable model—it’s also a more heavily governed one. As Anthropic regards Mythos-class models as having a higher risk profile than earlier Claude generations, Fable 5 comes with stricter safeguards, fixed retention rules, and clearer limits. This makes governance part of the product itself, not just something teams add on afterward.

Safety classifiers and fallback routing

Fable 5 sits behind a set of safety classifiers that watch for potentially harmful prompts and jailbreak attempts. When the classifiers decide a request is too sensitive—for example, a detailed cyber attack, dangerous biological information, or instructions that “distill” harmful knowledge—Fable 5 does not answer directly.

Instead, those requests are either blocked or routed to Claude Opus 4.8, which is still capable but has a smaller risk profile. According to Anthropic, this fallback should only occasionally trigger, with the safeguards mistakenly reacting to harmless requests in less than 5% of all sessions.

Data retention and auditability

Alongside the model‑level safeguards, Anthropic has changed how it handles logs for Fable 5 and Mythos‑class traffic. For these models, the company now keeps logs for at least 30 days across both its own platform and partner surfaces, even for customers that could previously negotiate shorter retention.

Anthropic says this data is not used to train future Claude models, but instead supports security and safety work, such as spotting complex, multi‑step attacks or reducing false positives from the safety classifiers. The logs themselves are covered by additional access controls and are deleted after the retention window in almost all cases. For regulated teams, this means Fable 5 can be monitored more closely, but it also means data‑handling policies need to account for the fixed 30‑day window.

Pricing and access

Anthropic prices both Claude Fable 5 and Claude Mythos 5 at 10 dollars per million input tokens and 50 dollars per million output tokens. While it says that the pricing is less than half the price of Claude Mythos Preview, it still places Fable 5 firmly in premium territory relative to lower Claude tiers.

Fable 5 is fully available now on the Claude API and consumption-based Enterprise plans like Pro, Max, and Teams. While it’s currently being offered to these plans at no extra cost, after June 23rd, Anthropic will remove Fable and require usage credits instead.

Where should you apply Fable 5?

Fable 5 would likely work best for workflows where higher per-token costs can be offset by higher autonomy, fewer failed attempts, and faster completion on work that would otherwise consume expensive human time.

Some good fits for Fable 5 include:

• Large codebase migrations and refactors, where the cost of human engineering time dominates model spend.
• Analytics and finance workflows that involve long document chains, charts, tables, and multi-step reasoning.
• Document-heavy knowledge work, especially when multimodal interpretation and long context matter.
• Agentic workflows that benefit from self-verification and fewer correction turns.

Where cheaper models may be better:

• Lightweight drafting and summarization.
• Basic support workflows.
• High-volume, low-complexity tasks where throughput matters more than peak reasoning quality.

While Fable 5 is the frontier option for hard problems, lower-cost Claude tiers remain a better fit for routine work that doesn’t need Mythos-class reasoning or long-horizon execution.

What Fable 5 means for the future

While Anthropic is cautious about Fable 5 and has gone to lengths to build safeguards, the company is also ultimately optimistic about it, Mythos 5, and the long-term benefits of both models.

“The capabilities of models like Fable 5 and Mythos 5 have the potential to do profound good for the world,” says Anthropic. “We’ve seen the beginnings of this in Project Glasswing, where the models have helped cyber defenders secure critically important software. We’ve also seen it in life sciences research, where the models are…speeding up the development of new therapeutics.”

For organizations looking at Fable 5 right now, the challenge is to turn that potential into something concrete and manageable. Teams need to decide which problems genuinely benefit from Mythos‑level capability, how to keep that use within security and compliance boundaries, and how to tell whether the extra spend is actually delivering better outcomes.

If your team is starting to experiment with Fable 5—or deciding how it should sit alongside the rest of the Claude family—it helps to have a plan. FullStack can work with you to pinpoint the specific workflows where a Mythos‑class model is worth the cost, set clear rules for how and when it’s used, and design the supporting processes, so Fable 5 becomes a dependable part of your production stack rather than an isolated experiment.

Contact us today to book a free consultation to understand where Fable 5 could fit in your business.