ReSearch: Learning to Reason with Search for LLMs via Reinforcement Learning

TL;DR

ReSearch introduces a reinforcement learning framework enabling large language models to effectively integrate search operations into multi-hop reasoning tasks without supervised reasoning data, enhancing their reasoning capabilities and generalizability.

Contribution

The paper presents a novel reinforcement learning approach that trains LLMs to perform search-based reasoning without supervised step data, improving reasoning and generalization.

Findings

Models demonstrate strong performance across multiple benchmarks.

ReSearch naturally elicits reflection and self-correction in reasoning.

Effective training on a single dataset with broad applicability.

Abstract

Large Language Models (LLMs) have shown remarkable capabilities in reasoning, exemplified by the success of OpenAI-o1 and DeepSeek-R1. However, integrating reasoning with external search processes remains challenging, especially for complex multi-hop questions requiring multiple retrieval steps. We propose ReSearch, a novel framework that trains LLMs to Reason with Search via reinforcement learning without using any supervised data on reasoning steps. Our approach treats search operations as integral components of the reasoning chain, where when and how to perform searches is guided by text-based thinking, and search results subsequently influence further reasoning. We train ReSearch on Qwen2.5-7B(-Instruct) and Qwen2.5-32B(-Instruct) models and conduct extensive experiments. Despite being trained on only one dataset, our models demonstrate strong generalizability across various benchmarks. Analysis reveals that ReSearch naturally elicits advanced reasoning capabilities such as reflection and self-correction during the reinforcement learning process.