Artificial Analysis Launches PersonalityBench and Finds Meta's Muse Models the Most Friendly
In AA-PersonalityBench, Meta AI’s Muse models are the most Friendly of all models The Friendly trait measures whether a response conveys warmth and approachability without feeling performative. ➤ Most Friendly: Muse Spark 1.3 (100), Muse Spark Open (98), Muse Wonder (95) and Muse Glimmer (93). Congratulations @ alex wang! ➤ Least Friendly: Nemotron 3 Ultra (1), GPT-3.5 Turbo (3), MiniMax-M3.1-Flash (6), Grok 4.6 (8) and DeepSeek V4.1 Flash 1017 (11)
Context
Benchmarks usually measure what models can do, not how they come across, so a personality score gives buyers a new way to compare assistants for customer facing work.
Artificial Analysis is an independent firm that benchmarks AI models on intelligence, speed and price. Muse is the family of models from Meta Superintelligence Labs, led by Alexandr Wang. AA PersonalityBench is a new benchmark that scores models on personality traits rather than task accuracy.
Deep dive
Artificial Analysis said on Wednesday that in its AA PersonalityBench, Meta's Muse models are the most Friendly of all the models it tested. It defines the Friendly trait as whether a response conveys warmth and approachability without feeling performative.
According to the post, the friendliest models are Muse Spark 1.3 with a score of 100, Muse Spark Open with 98, Muse Wonder with 95 and Muse Glimmer with 93. The least friendly are Nemotron 3 Ultra with 1, GPT 3.5 Turbo with 3, MiniMax M3.1 Flash with 6, Grok 4.6 with 8 and DeepSeek V4.1 Flash with 11. The chart shows the scores as bars across dozens of models.
The post did not explain in detail how responses are judged, whether by people or by other models, or how the scores are scaled, so the numbers are best read as rankings within this benchmark. A friendlier tone is also not the same as better answers, and some users prefer blunt, concise replies.
The result is a small win for Meta, whose Muse models have been trailing rivals on many capability rankings.
What to expect next
Watch whether Artificial Analysis publishes its full method and results for other traits, and whether model makers start tuning for these scores. Personality benchmarks could influence which models companies pick for support agents and companion apps. Critics will want to see that friendliness does not come at the cost of honesty or the flattering behavior labs have been trying to reduce.