Anthropic ships Claude Fable 5 — and an unlocked sibling, Mythos 5.

Two models, one brain

The headline is a pair. Fable and Mythos are the same model; the difference is who's allowed to use it and with which guardrails on.

Mythos 5, up close

The restricted sibling deserves its own paragraph, because the science claims attached to it are the most striking part of the announcement.

• Who has it today — Project Glasswing participants: cyber defenders and infrastructure providers. A broader trusted-access program is in development, including a dedicated biology track for biomedical researchers (cyber safeguards kept on, bio/chem protections lifted), launching in the coming weeks. A formal application process is planned but not yet published.
• Drug design, ~10× faster — Anthropic's internal protein-design experts report roughly tenfold acceleration. Of 14 protein targets tested, 9 produced strong drug-design candidates now under investigation — spanning immune checkpoints, growth-factor signaling, neurodegeneration, and muscle disease.
• Hypotheses scientists actually prefer — in blinded comparisons, scientists chose Mythos 5's molecular-biology hypotheses over Opus-class output about 80% of the time. One hypothesis about an E. coli protein mechanism was independently corroborated in peer research.
• A week of autonomous genomics — over one unattended week, the model assembled single-cell data across 138 animal species and built a custom ML model that outperformed a recent Science publication while being 100× smaller.
• Why it's locked up — Anthropic says Mythos-class capability presents "substantial risk of uplift to malicious actors" in cyber and biology; the same model that designs proteins can predict viral shell assembly. That dual-use reality is the whole reason the Fable/Mythos split exists.

Want the practical version — what Mythos means for you and how to start using the Mythos-class model that is available? Read the getting-started guide to Mythos & Fable.

What Anthropic claims it can do

• State-of-the-art on nearly all tested benchmarks — Anthropic singles out software engineering, knowledge work, vision, and scientific research.
• Long-horizon autonomy — works on its own across millions of tokens in long-context tasks, the pitch being sustained multi-day project work rather than one-shot answers.
• Real-world coding — Stripe reported Fable 5 "compressed months of engineering into days" on a Ruby codebase migration.
• Frontier science (Mythos 5) — a reported 10× acceleration on drug-design tasks, and Anthropic's claim that it's the first model to consistently produce novel, compelling scientific hypotheses in molecular biology.

Fable 5 compressed months of engineering into days on our Ruby codebase migration. — Stripe, as cited in Anthropic's announcement

The numbers

On safety

"Mythos-class" is the part worth reading slowly. The whole design of Fable 5 is that a frontier-capability model gets shipped to the general public only with classifiers in front of it — narrow domains (cyber, bio/chem, distillation) get bumped down to Opus 4.8 rather than answered at full strength. Anthropic reports external red-teaming found no successful single-turn jailbreaks across 30 published techniques. Mythos 5, the unguarded version, stays behind partner vetting precisely because those guardrails are off.

What it means for the WholeTech network

Source: Anthropic, "Claude Fable 5 and Claude Mythos 5" (June 9, 2026). This is a condensed summary; figures and quotes are as reported in the official announcement. See also the full Claude guide.