Fable 5.1's August window has closed — the market now prices September at 78%, and one claim in the leak can't be checked at all

A claim going round today says Fable 5.1 is in pre-release testing, targeted at September, and running inside Anthropic's own products under the Fable 5 name. One part of that is checkable, one part is checkable and slipping, and one part is — by construction — invisible to anything we can measure.

The market number is real, and it is worth reading properly. Polymarket carries "Next Mythos-Class Model released by…?", which is the instrument for this question even though its title contains neither "Fable" nor "Claude". Read at 15:55 UTC today: **July 31 has settled at 0%, August 31 trades at 25% (27% midpoint), September 30 at 77% (78% midpoint), and September 15 at 51%**. Volume on the event is about $138,000.

So the widely-quoted "78% for September" is the September 30 midpoint, and it is accurate. The more informative pair is the two numbers either side of it. **August 31 at 25% means the market has already largely written off the month the original leak named** — and September 15 at 51% says the weight inside September sits late rather than early. A leak that said August is now a market that says probably-not-August.

Kalshi agrees from the other side. Its next-Mythos-class contract — whose rules explicitly cover any model branded "Mythos" or "Fable", excluding Fable 5 — has now settled NO for both the August 1 and **August 14** windows. Three remain open: September 1, October 1, November 1.

Where the leak came from, and how far it has moved. Fable 5.1 first surfaced on July 26 in posts from community trackers, claiming an August window, long-horizon reasoning and agent work, and pricing held at Fable 5's $10/$50 per million. Anthropic's model documentation, product pages and newsroom have carried no mention of a Fable 5.1 at any point since. Four weeks later the target has moved a month. That is the ordinary shape of these things, and it is why we log dates rather than repeat them.

Nothing is in the catalogs: no fable-5.1 across 417 OpenRouter ids or 6,857 on models.dev. And the alias worth watching has not moved — `claude-fable-latest` still resolves to Fable 5's economics at $10/$50 with a 1M context, listed June 9 and unchanged. When 5.1 is real, that alias repointing is likely to be the first observable moment, ahead of any announcement.

Now the part we cannot check, and should say so rather than dress up. The claim that 5.1 is already serving inside Anthropic products under the Fable 5 name is not falsifiable from outside. A model swapped in behind an existing name ships **no new id**, so a catalog sweep sees exactly nothing — which is the same blind spot that let OpenAI's Ultrafast tier past us on August 13. We are not calling it false. We are saying plainly that no public evidence could confirm it, and that a claim built that way should be held differently from one a catalog can settle.

Two smaller things worth flagging. "Anthropic always tests before a release" is true of every lab and predicts nothing — it is the kind of line that makes a rumor feel evidenced without adding evidence. And the framing of 5.1 as a response to GPT-Astra sits awkwardly next to the record: OpenAI said on August 7 that it cannot rule out Critical cyber capability for Astra, is treating it as its first Critical model, and is **deliberately slowing it down**. A model positioned as the answer to a competitor's launch is being positioned against a launch that competitor has paused.

What we would actually watch, in order: the `claude-fable-latest` alias repointing; a non-variant fable-5.1 id appearing in any catalog; and the September 15 contract, which is the cheapest available read on whether the market thinks this lands early or late in the month. Our alert for Fable 5.1 has been armed since well before this leak and is unchanged.