Anthropic's Claude Fable 5.1 Reportedly Solves 1653 Royalist Cipher in 44 Minutes

What happened

Anthropic's Claude Fable 5.1 has reportedly solved a 373-year-old numeric cipher that had resisted attempts by other frontier AI models, according to testing organization Vals AI. The cipher, known as the "Cyphral Distich," was published in 1653 by Scottish writer Sir Thomas Urquhart and consists of two lines of 32 numbers each.

Vals AI says Fable 5.1 solved the puzzle in 44 minutes with no human intervention. The researcher behind the test had spent months running other leading models against a set of unsolved puzzles; none produced a verifiable solution before Fable 5.1's result.

How the model found it

According to Vals AI, Fable 5.1 was not given the Cyphral Distich directly. Instead, it was tasked with searching for a solvable puzzle on its own, sifted through multiple candidates, and flagged the Distich as promising.

The solution mechanism: each of the 64 numbers corresponds to a word located in one of the 32 numbered sections of Urquhart's original publication. Taking the first letter of each identified word spells out a coherent sentence: "O God uphold King Charls the Second and make him the supreme ruler of this land." The message reflects a pro-royalist sentiment consistent with the political climate of 1653, during the English Interregnum under Oliver Cromwell.

Vals AI characterizes the breakthrough as a product of systematic trial and error and persistence rather than superior cryptanalytic technique. The organization notes that, in retrospect, the solution method is straightforward enough that a human researcher could plausibly have solved it — the puzzle had simply gone unattempted at this level of rigor.

What we don't know

The source material does not disclose Fable 5.1's underlying architecture, parameter count, context window, or pricing. There is no independent verification beyond Vals AI's account, and no peer-reviewed confirmation that Urquhart's 1653 text is the sole valid key to the cipher. Claims about the model's search process and 44-minute solve time originate entirely from Vals AI's reporting and have not been independently reproduced by other researchers or publications.

What this means

This result is presented as a capability demonstration rather than a formal model release with documented specs — no context window, pricing, or benchmark suite accompanies the announcement. If accurate, it illustrates a genuine use case for long-horizon, self-directed search: an AI system choosing its own problem, iterating without human guidance, and arriving at a verifiable answer. That distinguishes it from typical benchmark scores, which measure performance on pre-defined tasks rather than autonomous problem selection.

The caveat matters: Vals AI is the sole source, and the claim rests on that organization's account of the model's process. Historical cryptography puzzles are also unusually forgiving test cases — a solution is either verifiably correct or it isn't, with little room for ambiguity, which makes them appealing demos but not necessarily representative of broader reasoning capability. Readers should treat this as an interesting anecdote about persistence-driven search until independently replicated.