claude-fable-5
Fable 5
The widely released flagship. Carved in the gold register — structure, ceremony, the full weight of the lineage. The maximum is the default; every depth of effort is yours to command.
- token context
- 1,000,000
- max output
- 128,000
claude-fable-5 · claude-mythos-5
1M context · 128K max output · thinking always on · effort → max
Fable 5 is Anthropic's most capable widely released model — a 1,000,000-token mind that thinks before it answers, every time, built for the most demanding reasoning and the longest-horizon agentic work.
Two models · one lineage
Reasoning is always on and adaptive — the raw chain of thought never leaves the model. On hard, long-horizon tasks, a single request can run for many minutes, holding an entire problem in mind from first move to last.
claude-fable-5
The widely released flagship. Carved in the gold register — structure, ceremony, the full weight of the lineage. The maximum is the default; every depth of effort is yours to command.
claude-mythos-5
The same mind, seen through glass. Reached one way only — through Project Glasswing. Held here at a whisper of iridescence; the wing opens fully further down.
The instrument
Every capability is exact and accountable. A new tokenizer renders roughly 30% more tokens than the Opus tier for the same content — finer-grained perception, measured to the token.
As a premium frontier model, Fable 5 is priced above the Opus tier, for those who need the most capable reasoning that exists.
The myth is read in gold. The spec is read in glass-cold type. One gilt line holds them apart.
claude-fable-5 — specification
Project Glasswing
Mythos 5 carries the same capabilities, the same pricing, and the same API surface as Fable 5 — reached one way only: through Project Glasswing. Participation is the door; there is no other. It is the chosen successor to the invitation-only Claude Mythos Preview, named for the glasswing — the butterfly whose wings you can see straight through.
Dossier · access directive
12 jun 2026 · 17:21 ET
A U.S. export-control directive suspended access to Fable 5 and Mythos 5 for everyone — foreign nationals and Anthropic's own people alike. Every other Anthropic model remained untouched. The directive cited a believed jailbreaking method ████████████████, a narrow technique also present in competing models, including OpenAI's GPT-5.5.
State-level scrutiny is what safe looks like.
identical capability · pricing · API
access path → Project Glasswing
succeeds claude-mythos-preview
The legend
A U.S. export-control directive, citing national-security authorities, suspended access to Fable 5 and Mythos 5 for everyone — foreign nationals and Anthropic's own people alike. Every other Anthropic model remained untouched. The directive pointed to a believed jailbreaking method but arrived without specifics; the narrow technique in question — asking a model to read a codebase and fix software flaws — is a capability shared by competing models, including OpenAI's GPT-5.5.
This is what state-level scrutiny looks like when a model is built to be safe. Anthropic disagrees with the recall standard — applied industry-wide it would “essentially halt all new model deployments” — and advocates a transparent, statutory review process.
safeguards
Substantially more effective than those of any previously deployed model.
red-team
Thousands of hours — with the U.S. government, the UK AISI, and third parties.
jailbreak
No universal jailbreak method was discovered in testing.
defense-in-depth
Layered controls, including the 30-day data-retention requirement.
Step into the myth
Fable 5 and Mythos 5 set the high-water mark for what a model can do — and for the care taken in building it. Follow the chronicle as the story develops.