Anthropic released Claude Fable 5 and Claude Mythos 5 on June 9, 2026 — the first generally-available models in its new Mythos class, a tier the company positions above Opus in raw capability. Fable 5 is the version made safe for general use; Mythos 5 is the same underlying model with some safeguards lifted, available only to a small set of vetted partners.
The names share an origin: fabula, Latin for "that which is told," and mythos, its Greek counterpart. The split between them is not about power — they are the same model — but about who is allowed to use it without guardrails.
What Fable 5 can do
Anthropic says Fable 5 posts state-of-the-art results on nearly every capability benchmark it tested, with its lead over earlier models widening as tasks get longer and more complex. It is strongest in software engineering, knowledge work, vision, and scientific research, and can hold focus across millions of tokens in long-running, autonomous work — longer, the company says, than any prior Claude.
Early third-party results back the claims:
| Test / partner | Result |
|---|---|
| Stripe — 50M-line Ruby migration | Compressed months of engineering into days |
| Cognition — FrontierCode eval | Highest score among frontier models |
| Hex — core analytics benchmark | First model to reach 90% |
| Base44 | Superior at "one-shotting" full apps |
| Genspark | Beat competitors on UI design and game coding |
The through-line is autonomy on hard, multi-step work — the kind that used to need a person checking each step.
The safety model: a fallback to Opus 4.8
The headline isn't only capability; it's restraint. Fable 5 ships with deliberately conservative safeguards. When a query touches a high-risk domain — cybersecurity, biology, chemistry, or model distillation — Fable 5 declines, and the request is instead answered by Claude Opus 4.8. Anthropic says these safeguards trigger in fewer than 5% of sessions on average, so at least 95% of conversations run entirely on Fable.
The cyber safeguards were stress-tested unusually hard: external red-teamers reported zero successful single-turn jailbreaks across 30 public techniques, and a bug bounty produced no universal jailbreaks in more than 1,000 hours of testing. Mythos-class traffic carries a 30-day data-retention policy and, Anthropic says, is not used to train models.
What Mythos 5 is — and who gets it
Mythos 5 is Fable 5 with some of those guardrails removed, released narrowly:
- Project Glasswing cyber defenders and critical-infrastructure providers — a group expanded last week to hundreds of organizations across 15 countries;
- selected biomedical researchers, who get the biology safeguards lifted while the cybersecurity restrictions stay in place.
The caution is not theoretical: in one evaluation, Mythos 5 outperformed specialized protein models at designing adeno-associated viruses — exactly the kind of dual-use capability that explains why the unrestricted model isn't public.
Pricing and availability
Both models cost $10 per million input tokens and $50 per million output tokens via the API — double Opus 4.8's $5 / $25, and less than half what the earlier Mythos Preview cost.
Fable 5 is available immediately through the Claude API and consumption-based Enterprise plans. On Pro, Max, Team, and seat-based Enterprise plans it's included at no extra cost from June 9 through June 22; from June 23 it shifts to usage credits until Anthropic adds enough capacity to fold it back into standard plans. Mythos 5 remains restricted to authorized partners.
Why it matters for working professionals
The release lands days after Anthropic publicly warned that frontier AI — particularly recursive self-improvement — is becoming dangerous, and shortly before the company's planned IPO. Fable 5 is its attempt to square that circle: ship its most capable model widely, but wrap it in a fallback that quietly routes the riskiest fraction of questions to a tamer model.
For most professionals, the practical story is the capability jump on long, complex tasks — a lawyer running a model across an entire matter, an engineer handing off a large migration, an analyst pushing through a dense dataset without losing the thread. The safety design is mostly invisible: unless your work touches cybersecurity, biology, chemistry, or model-copying, you'll probably never see the Opus 4.8 handoff.
The catch is cost and timing. At twice Opus pricing, Fable 5 is a premium tier, and the June 23 move to usage credits means the "free on your plan" window is only two weeks. For everyday drafting and analysis, Opus 4.8 likely remains the better value; Fable 5 earns its price on the long, autonomous jobs that smaller models still fumble.
Sources: Anthropic · TechCrunch · CNBC · NBC News