✦Dispatch · 11 June 2026
The frontier now ships in two editions.
Anthropic released Claude Fable 5 — the first model in the Claude 5 family, and the first of a new Mythos-class tier that sits above Opus. Alongside it: Claude Mythos 5, the same underlying model with fewer guardrails, available only to approved organizations. One model, two doors.
✦What shipped
A tier above Opus.
Fable 5 is pitched at the work prior frontier models couldn’t hold: long-horizon agentic runs where a single request can take minutes, parallel sub-agent delegation, file-based memory across sessions. Thinking is always on and can’t be turned off — and the raw chain of thought is never returned, only summaries. A new tokenizer spends roughly 30% more tokens on the same text, and the price is double Opus on both ends.
model access price (per MTok in/out) ───────────────────────────────────────────────────────────────────────── claude-fable-5 generally available $10 / $50 claude-mythos-5 approved orgs (Glasswing) $10 / $50 claude-opus-4-8 generally available $5 / $25 context window: 1M tokens (default) · max output: 128K
✦One model, two doors
Capability is no longer the only axis.
Fable 5 and Mythos 5 share the same weights. The difference is policy: Fable ships with additional safety measures for dual-use capabilities — classifiers that can decline a request mid-stream in domains like biology and cybersecurity — while Mythos 5 is offered without them, to organizations Anthropic has individually approved through a program called Project Glasswing.
That’s new. Frontier releases used to be segmented by capability and price. This one is segmented by trust— who you are determines which edition of the model you may hold. There’s one more condition worth noticing: Fable 5 requires 30-day data retention. It is not available under zero-data-retention agreements at all. The most capable model on the market is also the one you cannot use without leaving a copy of your prompts behind.
✦How we’re reading this
Rented intelligence, now with terms.
Fable 5 is genuinely remarkable, and nothing that runs on a laptop today comes close. But look at the shape of the deal: reasoning you can’t inspect, classifiers you can’t appeal, retention you can’t opt out of, and a stronger edition reserved for the approved. Every one of those terms is a reminder that frontier intelligence is something you rent, on the landlord’s conditions.
That’s not a criticism of Anthropic — at the frontier, those terms may even be the responsible ones. It’s the reason we’re building the other layer: intelligence you own, on your device, with no terms attached. The frontier and the floor are both rising. We work the floor.
✦ ✦ ✦
The ceiling rose again.
So did the rent.