2 Related Work

本文为论文《Search-o1：让推理模型学会边思考边检索》「论文原文」板块的对应章节，按 arXiv 官方全文逐字保留，层级与「中文全文翻译」板块一一对应。

Large Reasoning Models.

Large reasoning models focus on enhancing performance at test time by utilizing extended reasoning steps, contrasting with traditional large pre-trained models that achieve scalability during training by increasing model size or expanding training data [17, 66, 50, 85, 76]. Studies have shown that test-time scaling can improve the reasoning abilities of smaller models on complex tasks [15, 75]. Recently, models like OpenAI-o1 [22], Qwen-QwQ [54] and DeepSeek-R1 [7] explicitly demonstrate chain-of-thought reasoning [59], mimicking human problem-solving approaches in domains such as mathematics, coding, and so on.

Various approaches have been explored to achieve o1-like reasoning capabilities. Some methods combine policy and reward models with Monte Carlo Tree Search (MCTS) [25], though this does not internalize reasoning within the model. Other studies incorporate deliberate errors in reasoning paths during training to partially internalize these abilities [49, 71]. Additionally, distilling training data has been shown to enhance models’ o1-like reasoning skills [45]. The o1-like reasoning paradigm has demonstrated strong performance across diverse domains, including vision-language reasoning [65, 11, 48, 69], code generation [81, 32], healthcare [3], and machine translation [57]. However, these approaches are limited by their reliance on static, parameterized models, which cannot leverage external world knowledge when internal knowledge is insufficient.

Retrieval-Augmented Generation.

Retrieval-augmented generation (RAG) introduces retrieval mechanisms to address the limitations of static parameters in generative models, allowing access to external knowledge to solve more complex problems [30, 82, 35, 86]. Advanced research in this field enhances the RAG system from multiple aspects, including the necessity of retrieval [53], pre-processing of queries [43, 58], retrieved documents compressing [64], denoising [42, 12], refining [24, 27, 88], instruction following [9, 8, 87] and so on. Furthermore, some studies have explored end-to-end model training to implement RAG systems [1, 36, 33, 34] and knowledge-graph-based RAG systems [14, 37].

Recently, agentic RAG systems empower models to autonomously determine when and what knowledge to retrieve as needed, showcasing enhanced planning and problem-solving capabilities [5, 56, 70]. There is also research combining agent-based systems with MCTS to optimize complex workflows, leveraging retrievers and other tools to accomplish tasks [78]. However, existing RAG approaches have not combined the strong reasoning capabilities of o1-like models, limiting the potential to further enhance system performance in solving complex tasks.

Figure 2: Comparison of reasoning approaches: (a) Direct reasoning without retrieval often results in inaccuracies due to missing knowledge. (b) Our agentic retrieval-augmented reasoning approach improves knowledge access but usually returns lengthy, redundant documents, disrupting coherent reasoning. (c) Our Search-o1 integrates concise and accurate retrieved knowledge seamlessly into the reasoning process, enabling precise and coherent problem-solving.