Artificial Analysis Launches AA-PersonalityBench Scoring 30 Model Personality Traits

ArtificialAnlys · x · 2026-11-04

Artificial Analysis released AA-PersonalityBench, a benchmark measuring how models communicate rather than whether they are correct.

Method: 30 personality traits are judged pairwise — an LLM judge panel compares two answers to the same prompt and picks the one showing more of the trait, desirable or not. Each trait has 30 prompts (900 total), and verdicts become a Bradley-Terry Elo rating per trait.
8 indexes: Expressiveness, Organization, Informativeness, Transparency, Warmth, Conviction, Imagination and Slop, each the mean of its traits' Elo z-scores, percentile-ranked 1–100 across all evaluated models.
21 communication characteristics: length, structure, punctuation and emphasis measured directly from responses.
Leaderboard: top proprietary models include Claude Opus 5.5, GPT-6 Astra, GPT-6.1 Sol, Claude Sonnet 5.5 (max) and Gemini 4 Argon; top open-weights entries include Kimi K3.1 Pro. Results are group averages and individual models vary within groups.
A customizable personality ranking page and full methodology are published.

More from Models