Atlas

Benchmarks

← All benchmarks

LLM Divergent Thinking Repeat Rate

Chat & Writing · 2025-03-20

The Divergent Thinking benchmark reports a repeated-word rate as a diagnostic for output diversity. Lower repeat rates indicate fewer duplicated words.

Top models (lower is better)

ModelScore
Llama 3.3 70B Instruct0.0
o1 Preview0.0
Gemini 2.0 Flash Thinking Experimental 01-210.2
Claude 3.5 Sonnet (Oct 2024)0.5
Gemini 1.5 Flash0.7
Gemini 1.5 Pro 0020.9
Llama 3.1 405B1.0
Gemma 2 27B1.1
o1-mini2.2
Gemini 2.0 Flash Experimental2.6
Opus 34.0
Mistral Large 2 (Instruct 2407)5.0
Haiku 3.55.5
Grok 25.6
GPT-4o Mini8.4
Qwen2.5 72B Instruct11.2
Claude 3 Haiku16.4
GPT-4o23.7
DeepSeek-V2.525.1
Loading Atlas data…