返回全部资讯
Artificial Analysis 发布 AA-PersonalityBench 量化 AI 个性与沟通风格Artificial Analysis (@ArtificialAnlys)T1·基座大模型
AA-PersonalityBench 显示低智模型更易产出 Slop
原标题：In AA-PersonalityBench, less intelligent models generally produce more Slop, but Slop still varies w...
45AI 研判分
Artificial Analysis 发布 AA-PersonalityBench 基准测试，指出模型智力水平与“Slop”（低质/废话）生成量总体呈负相关。然而，前沿模型间差异巨大：GPT-6 Sol 和 Muse Spark 1.3 虽智力指数相同，但 Slop 排名分别为第 1 和第 80，表明高智力并不必然抑制低质输出。
报道正文
Artificial Analysis (@ArtificialAnlys) · 2026-10-22 22
In AA-PersonalityBench, less intelligent models generally produce more Slop, but Slop still varies widely among frontier models
➤ Same intelligence level but opposing Slop rankings: GPT-6 Sol and Muse Spark 1.3 both score 48 on the Intelligence Index, yet rank 1 and 80 on Slop respectively
➤ High intelligence but also high on slop: Muse Spark 1.3 (Slop 80), Qwen3.8 Max (75) and Gemini 4 Argon (70)
➤ Low slop for their intelligence level: GPT-6 Luna (4) and Claude Haiku 5.5 (9)
