Artificial Analysis PersonalityBench (AA-PersonalityBench)
We show this table for reference; we do not rank on it.
Measures how models write by default across 30 personality traits and 21 communication characteristics. A stronger trait is not necessarily a better answer.
Evaluation results published by Artificial Analysis.
How to read this benchmark
Use the source results to compare writing and communication preferences. Percentile ranks describe the source cohort and do not measure correctness, general intelligence, or an overall best personality.
Operator receipt: The original methodology and results are linked; this page provides a source reference.
Honest limit: The 900-question set is private, results depend on the judge panel and tested model settings, and cohort changes can move percentile ranks. The catalog links the original results without republishing its numerical leaderboard. These style measurements do not enter overall or category rankings.
How we show AA-PersonalityBench
AA-PersonalityBench compares default writing style across 30 judged traits, eight personality indexes, and 21 measured communication characteristics. Each model answers the same 900 questions without a system prompt.
These results describe how strongly a trait appears. We link the original results and methodology, keep them outside weighted rankings, and do not calculate an overall personality score.
Snapshot
About AA-PersonalityBench
Year
2026
Tasks
900 questions: 30 per judged trait
Format
30 trait Elo ratings, eight percentile indexes, and 21 communication characteristics
Difficulty
Default writing and communication style
Each model answers 900 questions once, as single user messages with no system prompt. A blinded, order-randomized panel compares answers on 30 traits. Firth-penalized Bradley-Terry ratings produce trait Elo scores; eight indexes average trait z-scores and report percentile ranks across evaluated models. Sycophancy is inverted only in the Transparency index. Another 21 communication characteristics are measured from the same responses.
Freshness and provenance
Version
AA-PersonalityBench 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does AA-PersonalityBench measure?
AA-PersonalityBench describes default writing and communication style through 30 judged traits, eight percentile indexes, and 21 measured communication characteristics. The same 900 questions are answered without a system prompt. Stronger expression of a trait does not establish better answers, correctness, or a generally better model.
Where are the personality results available?
Artificial Analysis publishes the numerical results on its AA-PersonalityBench evaluation page. We link that original source and explain its protocol here. This source-reference page does not reproduce the numerical table, calculate an overall personality score, or add these style measurements to overall and category rankings.
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 6,100+ subscribers.
One email each week. Unsubscribe anytime.