Artificial Analysis Launches PersonalityBench, Ranking AI Models by Personality Traits

Artificial Analysis has released a new benchmark, AA-PersonalityBench, aiming to quantify differences between models that are "equally correct but feel different to use": roughly 900 curated questions, with an LLM jury doing pairwise comparisons of answers to the same question (Reddit user likeastar20 relayed that the Elo scores come from the jury evaluating 30 prompt sets). Models are scored on 30 personality traits and 21 communication-style dimensions, rolled up into 8 personality indices: expressiveness, structure, informativeness, transparency, warmth, conviction, imagination, and Slop.

Confirmed

• Released by Artificial Analysis; the method is pairwise LLM-jury scoring covering 30 traits, 21 communication styles, and 8 aggregate indices.
• Sycophantic dimension: the three oldest models tested—GPT-4o, Llama 4, and others—rank as most sycophantic, while Grok 4.7 ranks last, i.e., least sycophantic / most honest.
• Information density dimension: Claude Fable 5.1 tops the list with a score of 100 (percentile), and the Claude family dominates both information density and prose quality.
• GPT-6 Astra's Slop score is only 16/100 (per a repost by @maxpaperclips), a result that has sparked controversy in the community.

Why it matters

• The benchmark gives users a new basis for choosing models by use case—e.g., creativity favors imagination, while technical work favors information density and honesty.
• The sycophancy results match the trend that newer models are less sycophantic, but using an LLM jury to score personality is itself debated, with GPT-6 Astra's low Slop score a focal point of controversy.

2026-10-21 ~ 2026-10-22 · 6 related posts

• AA-PersonalityBench ranks model sycophancy: GPT-4o tops the flattery chart, Grok 4.7 least — likeastar20 · 2026-10-21
• [source] AA-PersonalityBench: GPT-4o and Llama 4 top sycophancy ranking, Grok 4.7 lowest — ArtificialAnlys · 2026-10-22
• AA-PersonalityBench ranks model personalities: GPT-4o most sycophantic — ArtificialAnlys · 2026-10-22
• Artificial Analysis Launches PersonalityBench; GPT-6 Astra Scores 16/100 on Slop, Sparking Debate — max_paperclips · 2026-10-22
• [source] AA-PersonalityBench: Claude models sweep Information Density and Literacy rankings — ArtificialAnlys · 2026-10-22
• [source] Artificial Analysis launches AA-PersonalityBench: Claude models lead on information density and literacy — ArtificialAnlys · 2026-10-22