Muse Wonder vs. Claude Fable 5.1: Meta's New Model Came Closer Than I Expected
The New Stack, Friday, October 16th, 2026
A head-to-head test on agentic and memory tasks finds Meta's cheaper model close to Fable, but costlier in tokens.
The author ran Meta's Muse Wonder ($4/$20 per million tokens) against Anthropic's Claude Fable 5.1 ($10/$50) on three agentic and memory tasks graded by script.
On a debugging task, Fable passed all hidden tests while Muse missed one, and because Muse used about twice the input tokens, it cost more despite its lower prices.
Both scored 10 of 10 on an incident root-cause task, with Fable using fewer turns, while on a long-memory task Muse scored perfectly and Fable hit an API refusal flagged as cyber.
The author notes that the results come from single runs with default reasoning settings.