· 10 min
AI
The Best Model at the Thing It Won't Do
An independent evaluation gave Claude Fable 5 1,122 rare-disease cases. It refused 1,115. Opus 4.8, Opus 4.6, and GPT-5 refused zero. On the handful it could be coaxed into answering, it beat all three. That inversion — best when it speaks, silent most of the time — is the most absurd thing about Fable 5, and it breaks something benchmarks have always quietly assumed.