Prompt Repetition Improves Non-Reasoning LLMs

Leviathan, Yaniv, Kalman, Matan, Matias, Yossi (2025) 10.48550/arXiv.2512.14982

arXiv preprint arXiv:2512.14982

Brief Shows that repeating the input prompt improves performance for popular models (Gemini, GPT, Claude, Deepseek) without increasing the number of generated tokens or latency. Wins 47 out of 70 benchmark tests with 0 losses.

Abstract When not using reasoning, repeating the input prompt improves performance for popular models (Gemini, GPT, Claude, and Deepseek) without increasing the number of generated tokens or latency.

Keywords llm, prompt engineering, inference, performance

Dated: 4/26/26
Publish: 8/11/26
Revise: 8/14/26
Reading time: ~1 min read

Without reasoning, repetition improves performance

When not using reasoning, repeating the input prompt improves performance