Detailed "Muse Wonder 1.0" benchmark table leaks — every model in it is fictional
Every_Foundation5197 · reddit · 2026-10-13
A detailed "Muse Wonder 1.0 Benchmarks" table claims Meta's new Fable-class model tops every category: 59.1 on a "Personal Superintelligence Index" vs Claude Opus 5.5 (55.2) and GPT-6 Astra (54.5), plus leading scores on Harvey LAB-AA (13.1), SWE-Marathon (57.5) and MemoryBench (85.5).
The catch: every compared model — Claude Opus 5.5, Claude Fable 5.1, GPT-6 Astra — is fictional. It's a well-crafted fake benchmark sheet, interesting as AI-community lore rather than real news.
Related event: Meta Unveils Muse Wonder, Its Most Capable Frontier Model(92 posts)→
More from Fun
- A 1996 BBC clip on computer art and AARON, the pioneer of generative AI art — technollama · 2026-10-13
- Crimsonland Remake 0.15 Ships Custom Quest Language and Uncompressed 2004 Source Art — banteg · 2026-10-13
- After p-hacking, the AI crowd coins "O-hacking" for gaming benchmarks — chrisalbon · 2026-10-13
- Surreal AI-Generated 'Weird Buildings' Look Eerily Photorealistic — miilesus · 2026-10-13
- Using Claude in Cursor to Generate a History-of-Music Video — prasenx · 2026-10-13
- Rabbit Uses OS3 to Make a Chaotic Shitpost Meme About Itself — SimonBalmain · 2026-10-13