Artificial Analysis Ships AA-PersonalityBench, HealthBench Leaderboards

Artificial Analysis on October 19 published two new leaderboards that score frontier models on style and clinical conversation, not raw intelligence. AA-PersonalityBench grades 36 models on 30 personality traits that roll up into eight indexes — Expressiveness, Organization, Informativeness, Transparency, Warmth, Conviction, Imagination and Slop — alongside 21 communication characteristics measured directly from every response. The same day, the firm released independent HealthBench and HealthBench Professional rankings.

On the clinical side, "GPT-6 Astra leads HealthBench Professional at 66.0%" with OpenAI models occupying the entire cost-versus-performance frontier, while "Claude Sonnet 5.5 tops the broader HealthBench at 62.1%, narrowly ahead of DeepSeek V4.1 Flash at 60.9%." HealthBench Professional contains 525 physician-authored tasks across care consult, writing and documentation, and medical research, graded with GPT-5.4 as judge (GPT-4.1 grades the general HealthBench). Scores are reported raw and length-adjusted, with a penalty of 1.47 percentage points per 500 characters above or below 2,000.

Verbosity costs real ranking: "Claude Opus 5.5 with maximum reasoning and fallback earns the benchmark's highest raw score at 71.4%, then falls to sixth place with a 59.9% adjusted score." At the other extreme, "GPT-6 Sol produces the shortest answers at roughly 1,000 characters and leads communication quality with an 81% raw score. It also records the lowest completeness score at 34%."

The personality benchmark uses three LLM judges — Gemini 3.8 Flash, Claude Sonnet 5.5 and GLM 5.3 Flash — comparing responses pairwise across 900 questions, with results converted to Elo-style ratings using a Firth-penalized Bradley-Terry model and GPT-6.1 Sol anchored at 1,000 points. The style axis is new; the broader generative-AI tracker has carried 316 stories in the last ninety days that lean overwhelmingly on intelligence and cost scores.