← All benchmarks

AA-PersonalityBench Conviction index

Chat & Writing

Artificial Analysis's AA-PersonalityBench Conviction index, which measures direct, confident positions and constructive pushback. A panel of LLM judges compares two models' answers to the same prompt and picks the one showing more of a trait, whether or not the trait is desirable; verdicts become Bradley-Terry Elo ratings for 30 traits over 900 prompts. The index is the mean of its traits' Elo z-scores, percentile ranked from 1 to 100 across every model AA has evaluated, so a model's value moves as AA adds models. AA states it is not a capability benchmark: it describes how a model writes, not whether it is right, so it has no better direction and stays out of the capability score.

Benchmark source: Artificialanalysis

Top models (custom direction)