Artificial Analysis launches benchmark for AI model personality traits
Artificial Analysis introduced the AA Personality Bench to evaluate model behavior beyond factual correctness. The tool measures eight distinct indices to help users match specific AI models to targeted use cases.
Artificial Analysis announced the AA Personality Bench, a new evaluation framework designed to measure differences in model personality independent of answer accuracy. The benchmark assesses responses to 900 curated questions representing 30 personality traits, which are aggregated into eight indices: Expressiveness, Organization, Informativeness, Transparency, Warmth, Conviction, Imagination, and Slop. It also tracks specific communication characteristics, including sentence length, emoji usage, and em dash frequency.
Initial results show distinct performance profiles across leading models. Fable 5.1 and GLM 5.3 scored highest on the Expressiveness Index, while Muse Wonder, the model powering the Muse agent, led the Warmth Index. Opus 5-5 achieved the highest score on the Imagination Index, surpassing Fable 5-1 and GPT-6 Astra. Conversely, Gemini's 4 Argon model scored highest on the negative Slop Index. Artificial Analysis argues that personality traits are distinct from capability and are critical for matching models to specific user needs, such as selecting a model with high imagination for creative writing or one with high directness for legal applications. This development follows a broader industry trend where labs like Meta, OpenAI, and Anthropic are increasingly treating personality as a competitive differentiator, with Meta's Muse agent noted for its distinct character and mascot.