Atlas

Benchmarks

← All benchmarks

Slop Score

Chat & Writing · 2025-10-25

Slop Score measures how often generated prose uses stock phrases and other stylistic habits associated with LLM writing. Lower scores indicate less formulaic prose.

Top models (lower is better)

ModelScore
Kimi K2 Instruct 090518.3
Sonnet 4.519.5
Kimi K2 Instruct23.7
GPT-5 Mini26.3
Sonnet 427.3
o331.8
GPT-5 Nano34.1
ChatGPT-4o Latest (source-unspecified snapshot)47.7
Llama 4 Maverick Instruct49.2
Mistral Small 3.2 24B Instruct 250652.7
DeepSeek-V3.2-Exp54.2
Mistral NeMo Instruct 240754.4
Ling-1T56.2
GLM-4.559.9
Qwen3 4B62.2
DeepSeek-R167.9
Gemma 3 27B IT69.5
Gemma 3 12B IT70.6
Gemini 2.5 Flash77.6
Gemma 3 4B IT85.2
Loading Atlas data…