Artificial Analysis launched AA-PersonalityBench on October 19, 2026, using 900 questions covering 30 personality traits across eight indexes, along with 21 directly measurable characteristics. Answers are compared in blinded, randomized head-to-head evaluations and scored using a Firth-penalized Bradley-Terry Elo method combined with z-scores. Claude Fable 5.1, Opus 5.5, Sonnet 5.5, and Haiku 5.5 performed strongly in the two headline categories, generally ranking between the mid-90s and the 100th percentile. Claude Fable 5.1 scored 100 in Information Density and 95 in Literacy, while Claude Opus 5.5 received 97 in both. The benchmark measures communication style rather than overall capability, reinforcing Anthropic’s reputation for clear and substantive writing without establishing which model is the smartest.
Source: Read the original article
