Search-o1: Agentic Search-Enhanced Large Reasoning Models
Paper: arXiv:2501.05366 · Code: github.com/sunnynexus/Search-o1 · Released 2025-01-09
Authors: Xiaoxi Li, Guanting Dong, Jiajie Jin, Yuyao Zhang, Yujia Zhou, Yutao Zhu, et al. (8 authors)
Overview
Search-o1 enhances large reasoning models (LRMs) such as OpenAI-o1 with an agentic retrieval-augmented generation (RAG) mechanism and a Reason-in-Documents module for refining retrieved documents.
Motivation
LRMs demonstrate impressive long stepwise reasoning capabilities through large-scale reinforcement learning. However, their extended reasoning processes often suffer from knowledge insufficiency, leading to frequent uncertainties and potential errors. Search-o1 tackles this by letting the model fetch external knowledge on demand during reasoning.
Method
The framework has two key components:
1. Agentic search workflow — integrated into the reasoning process, enabling dynamic retrieval of external knowledge whenever the LRM encounters uncertain knowledge points. 2. Reason-in-Documents module — because retrieved documents are verbose, a separate module deeply analyzes the retrieved information before injecting it into the reasoning chain, minimizing noise and preserving a coherent reasoning flow.
Results
Extensive experiments on complex reasoning tasks in science, mathematics, and coding, as well as six open-domain QA benchmarks, demonstrate strong performance. The approach enhances the trustworthiness and applicability of LRMs in complex reasoning tasks.
Original Abstract (English)
> Large reasoning models (LRMs) like OpenAI-o1 have demonstrated impressive long stepwise reasoning capabilities through large-scale reinforcement learning. However, their extended reasoning processes often suffer from knowledge insufficiency, leading to frequent uncertainties and potential errors. To address this limitation, we introduce Search-o1, a framework that enhances LRMs with an agentic retrieval-augmented generation (RAG) mechanism and a Reason-in-Documents module for refining retrieved documents. Search-o1 integrates an agentic search workflow into the reasoning process, enabling dynamic retrieval of external knowledge when LRMs encounter uncertain knowledge points. Additionally, due to the verbose nature of retrieved documents, we design a separate Reason-in-Documents module to deeply analyze the retrieved information before injecting it into the reasoning chain, minimizing noise and preserving coherent reasoning flow. Extensive experiments on complex reasoning tasks in science, mathematics, and coding, as well as six open-domain QA benchmarks, demonstrate the strong performance of Search-o1. This approach enhances the trustworthiness and applicability of LRMs in complex reasoning tasks, paving the way for more reliable and versatile intelligent systems.
Engineering Takeaways
The forum discussion situates Search-o1 within the broader agentic search landscape and highlights practical considerations:
- Architecture: Agentic paradigms are turning retrieval count and strategy into learnable decisions, beyond the classic retrieve → rerank → generate cascade.
- Latency & cost: Iterative retrieval adds p99 latency; cascading with early stopping, caching, and budget control helps.
- Evaluation: Static nDCG is giving way to process metrics such as task success rate, citation accuracy, and multi-hop reasoning chain completeness; LLM-as-judge should be cross-validated with human evaluation.
- Safety: Open-web retrieval introduces poisoning and bias risks; source whitelisting and output filtering are advisable.
- Agentic Information Retrieval (arXiv:2410.09713)
- Synergizing RAG and Reasoning: A Systematic Review (arXiv:2504.15909)
- AceSearcher: Bootstrapping Reasoning and Search for LLMs via Reinforcement Learning