Artificial Analysis presented AA-PersonalityBench on October 19, 2026, a new benchmark that evaluates 38 AI models across 30 personality traits using 900 questions. Responses are scored by panels of large language models in blinded pairwise comparisons, then converted into Elo ratings, z-scores and percentiles. GPT-4o, in its November 2024 release, records the highest sycophancy score, while Grok 4.7 at the xhigh setting ranks last, with a gap of 775 Elo points. The benchmark’s eight indexes include transparency, warmth and slop, with Meta Llama standing out prominently in the latter. The framework measures communication style rather than raw model capability, but its rankings remain relative and may change as new models are added.
Source: Read the original article
