Trending storyDeveloping

Artificial Analysis launches AA-PersonalityBench for rating model personality and style

1 article1 sourcesince Oct 19Last article 34m ago · Oct 19, 2:27 PM ET

Overview

AISummary of 1 article

Artificial Analysis announces AA-PersonalityBench, a benchmark that scores language models on 30 personality traits and 21 communication characteristics.

The traits are rolled into eight indexes. The company says it evaluates models' answers to 900 curated questions, with subjective personality traits judged from those answers and objective communication characteristics calculated programmatically.

In the initial results posted on X, Claude Fable 5.1 leads four of the eight indexes, including Informativeness, Transparency, Warmth and Conviction. Qwen3.8 27B averages about 1,030 words per response, against 270 for GPT-6 Luna.

Written by AI from the articles below · updated Oct 19, 2:29 PM ET

Check the sources:

Latest developmentOct 19, 2:27 PM ET

AA-PersonalityBench measures traits and communication style across language models

Article timeline

The articles in this story. Times are ET.

Oct 19

2:27 PM
Artificial Analysis@ArtificialAnlysOfficial
AA-PersonalityBench measures traits and communication style across language models
AIArtificial Analysis announces AA-PersonalityBench, which scores language models on 30 personality traits rolled into eight indexes and on 21 communication characteristics. Claude Fable 5.1 leads four of the eight indexes, including Expressiveness-adjacent Informativeness, Transparency, Warmth and Conviction. Qwen3.8 27B averages about 1,030 words per response, nearly four times GPT-6 Luna's 270.

Heat Trend

Not enough continuous observations to show a trend yet.