What Is EQ-Bench Slop Score? (Explained Like You’re 5)
A quantitative metric created by EQ-Bench that measures how frequently an AI model repeats overused romance clichés, purple prose, and "Not X, but Y" rhetorical tics.
Have you ever read a cheap romance novel where every single page says "a shiver ran down her spine" or "his voice was barely a whisper"? The EQ-Bench Slop Score is a cliché radar that counts how many times per 1,000 words an AI bot falls back on lazy, robotic stock phrases.
How Major AI Companion Apps Compare on EQ-Bench Slop Score
The Exact Formula Behind EQ-Bench Slop Detection
In EQ-Bench’s open-source metrics.js, Slop Score combines three signals per 1,000 words: (1) 60% weight on single slop words ("tapestry", "testament", "unspoken", "palpable"), (2) 25% weight on "Not X, but Y" contrast patterns ("It wasn’t just coffee—it was a reckoning"), and (3) 15% weight on 3-word roleplay trigrams ("shiver ran down", "voice barely whisper", "breath caught throat").