The Slop Index

2026-10-11 · Docs and copy

What it measures: How chatbot-like a model's writing is across email, blog/essays, tweets, and Discord chat. Higher scores mean less chatbot-like writing.
How: Slashy's The Slop Index. Thirty-nine frontier models, 112 scenarios in four domains, compared with pre-2022 human writing. The blend is a live blind arena (40%) plus conciseness, templating, rhythm, and AI tells (15% each). The source Slop Score is 0–100, higher = worse, #1 = sloppiest; this site inverts so least slop is #1 and stores 100 − Slop Score. Open data is on GitHub.
What the score means: This site's score is 100 − the source Slop Score; higher is more human-like. The source ranks the highest slop as #1; ranks here are inverted so least slop is #1. Composite uses the pass-rate ratio to the leader; with one board, composite equals that normalized score.
Caveats: Arena votes are still accumulating, so the live Elo moves. Mechanical axes are normalized within this board. It does not measure code or UI quality.
Updates: Source https://theslopindex.com/leaderboard ; open data https://github.com/hgaddipati1118/slop-index . This page fetched all 39 rows on 2026-10-11 (1,795 votes then).
Source: theslopindex.com/leaderboard github.com/hgaddipati1118/slop-index

Full board

#	Model	Score