Claude Fable 5: Capabilities, Specs, Benchmarks & Why the US Locked It Down

On June 9, 2026, Anthropic released Claude Fable 5, the first publicly available model in its new Mythos class — a tier the company places above Opus. Three days later, the US government issued an export-control directive that forced Anthropic to suspend Fable 5 and Mythos 5 worldwide. This guide explains what Fable 5 actually is, how it performs, why it was banned, and what engineering teams should do now.

What is Claude Fable 5?

Claude Fable 5 is Anthropic's most capable generally available model. It shares the same underlying weights as Claude Mythos 5, but adds a safety layer that routes sensitive requests to the less-capable Claude Opus 4.8. Anthropic describes the Mythos class as its new top tier, sitting above Opus, Sonnet, and Haiku.

Benchmark performance

On Anthropic's launch benchmarks, Fable 5 reports state-of-the-art scores on coding and reasoning tasks. Independent verification is still pending, but the headline numbers are significantly ahead of Opus 4.8 and GPT-5.5 on software-engineering evaluations.

Real-world capability demonstrations

Beyond benchmarks, Anthropic and early partners published several high-signal production tests that show what the model can do at scale.

• Stripe used Fable 5 to complete a codebase-wide migration across a 50-million-line Ruby codebase in a single day — work the company estimated would take a team more than two months by hand.
• The model rebuilt the source code of a web application from screenshots alone, demonstrating strong vision-to-code capability.
• In a long-horizon test, Fable 5 completed Pokemon FireRed using only raw game screenshots as input.
• On scientific reasoning tasks, the unrestricted Mythos 5 version produced novel hypotheses in drug design and molecular biology, according to Anthropic.

Safety architecture: the Opus 4.8 fallback

Fable 5's defining architectural choice is its classifier system. Three classifiers scan every prompt for cybersecurity risk, biology or chemistry risk, and model-distillation attempts. If a prompt trips a classifier, the request is silently routed to Claude Opus 4.8 and the user is notified. Anthropic says this happens in fewer than 5% of sessions on average, meaning more than 95% of traffic runs on the full Mythos-class model.

The trade-off is frontier capability for most users, with a quiet downgrade for a small set of sensitive domains. For teams whose work repeatedly trips the filters, Anthropic offers a trusted-access path to Mythos 5 through Project Glasswing.

Pricing and availability before the ban

At launch, Fable 5 was available through the Claude API, Claude consumer and enterprise plans, Amazon Bedrock, Google Vertex AI, Microsoft Foundry, GitHub Copilot, and Databricks. It was free for Pro, Max, Team, and Enterprise users from June 9 to June 22, 2026, after which it would require usage credits.

The US ban: what happened and who is affected

On June 12, 2026, Anthropic announced that the US government had issued an export-control directive classifying Fable 5 and Mythos 5 as controlled technology. Because API keys cannot be nationality-checked, Anthropic suspended both models globally. Anthropic publicly disagreed with the scope of the order, stating that the cited security concerns involved a limited jailbreak that did not justify a blanket shutdown.

The practical impact is immediate: non-US developers and companies that had started piloting Fable 5 lost access overnight. Teams building products on the Claude API must now fall back to Opus 4.8, Sonnet, or Haiku, or evaluate alternatives such as Kimi K2.7 Code, GPT-5.5, or DeepSeek V4.

What this means for engineering teams

• If you are inside the US, Fable 5 access may resume under a trusted-access or nationality-verified program, but timelines are unclear.
• If you are outside the US, you should treat Fable 5 as unavailable and re-architect around models you can reliably access.
• For agentic coding, Kimi K2.7 Code offers comparable agentic benchmark performance at roughly one-fifth the API cost and is not subject to the same export restrictions.
• For general reasoning and long-context work, Claude Opus 4.8 and GPT-5.5 remain viable, though neither matches Fable 5's reported SWE-bench Pro score.
• For self-hosting and data sovereignty, open-weight models such as Kimi K2.7 Code and DeepSeek V4 are now the safest strategic bet.

Frequently asked questions

Is Claude Fable 5 the same as Claude Mythos 5?

They share the same base model. Fable 5 is the public version with safety classifiers; Mythos 5 is the restricted version with fewer guardrails, available only to vetted partners through Project Glasswing.

Why was Fable 5 banned?

The US government classified the models as controlled technology following reports of a jailbreak that could be used to bypass safety filters. Anthropic argues the cited risk was limited and does not justify broad restrictions.

Will Fable 5 come back?

Anthropic says it is working to restore access as quickly as possible, but no timeline has been provided. The most likely path is a nationality-verified or trusted-access program rather than fully open API access.

Need help choosing and integrating the right AI model for your Proptech or Fintech product? Book a call with the SelectCursor team and we'll help you evaluate open-weight and proprietary options against your compliance, cost, and accuracy requirements.