Frontier Design Benchmark - Slop Detector: leaderboard

How AI-made a model's websites look on the Frontier Design Benchmark, where models build working websites from real client briefs: a classifier trained on human-made and AI-made pages reads a few hundred visual and writing features of a page and reports the share of AI-made pages it looks more AI-like than (0-100, lower is less AI-looking), averaged over the landing pages a model built. It is a style reading, not a quality score: the human reference designs read near 0, and the publisher notes that one low-ranked model also reads near 0 because many of its pages are unfinished.

Metric: Slop Detector Reading (0-100, lower is less AI-looking). Source: contralabs.com. 16 models tracked. Updated 19 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1DeepSeek V4 Pro1.2#198
2Gestalt1.2
3Kimi K317.5#61
4GLM 5.3 Prime24.5#81
5Grok 4.725.5#51
6Claude Fable 5.231.4#1
7Claude Opus 5.531.5#9
8Qwen 3.8 Max31.6#66
9Muse Wonder33#34
10Muse Spark 1.335.4#48
11Claude Sonnet 5.537.7#24
12GPT-6 AstraUltra41.4#10
13GPT-6.1 Sol52.8#15
14Gemini 4 Carbon53.6#5
15Gemini 4 Argon57.3#18

Interactive version: theaggregate.ai/benchmark?slug=frontier-design-benchmark-slop-detector · How It Works · Data refreshed daily, snapshot 2026-11-20.