FULL STORY
AA-PersonalityBench: Benchmarking LLM Personalities
Artificial Analysis launched AA-PersonalityBench to measure how models communicate rather than whether they are correct. Follow-up group comparisons showed open-weight models trailing proprietary ones across eight personality indices.
2026-11-03 ~ 2026-11-04 · 2 episodes · 5 posts
Episode 1 · Artificial Analysis launches AA-PersonalityBench to score model personalities (2026-11-03, 3 posts)
Artificial Analysis released AA-PersonalityBench, an open benchmark scoring LLMs on 30 personality traits to measure how models communicate rather than correctness. Open-weight models trailed closed ones by 32 points on genuineness but produced less slop.
- AA-PersonalityBench goes live: build your own AI personality rankings across 41 models — ArtificialAnlys · 2026-11-03
- AA-PersonalityBench: open models trail proprietary by 32 points on truthfulness — ArtificialAnlys · 2026-11-04
- Artificial Analysis Launches AA-PersonalityBench Scoring 30 Model Personality Traits — ArtificialAnlys · 2026-11-04
Episode 2 · AA-PersonalityBench update: open models lag closed ones across all indices but slop (2026-11-04, 2 posts)
Artificial Analysis detailed the AA-PersonalityBench model lineup—Claude Opus 5.5 and GPT-6 versus Kimi K3.1 and DeepSeek V4.1 among others—and showed open-weight models trail proprietary ones on all 8 averaged personality indices except slop.
- Open models trail proprietary on every personality index except slop — ArtificialAnlys · 2026-11-04
- AA-PersonalityBench lineup: Opus 5.5 and GPT-6 vs Kimi K3.1 and DeepSeek V4.1 — ArtificialAnlys · 2026-11-04