Search-o1: Agentic Search-Enhanced Large Reasoning Models

Paper: Search-o1: Agentic Search-Enhanced Large Reasoning Models (arXiv, 2025-01-09) Authors: Xiaoxi Li, Guanting Dong, Jiajie Jin, Yuyao Zhang, Yujia Zhou, Yutao Zhu, et al. (8 authors) Code: https://github.com/sunnynexus/Search-o1

Overview

Large reasoning models (LRMs) like OpenAI-o1 demonstrate impressive long stepwise reasoning 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 with two components:

1. Agentic RAG mechanism — an agentic search workflow is integrated into the reasoning process, enabling dynamic retrieval of external knowledge whenever the LRM encounters uncertain knowledge points. 2. Reason-in-Documents module — since 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.

Key Contributions

• A framework that plugs agentic retrieval into LRM reasoning loops rather than using one-shot RAG.
• A Reason-in-Documents module that distills retrieved documents before they re-enter the reasoning chain.
• Extensive experiments on complex reasoning in science, mathematics, and coding, plus six open-domain QA benchmarks.

Evaluation

Per the paper's abstract, experiments cover:

• Tasks: complex reasoning (science, mathematics, coding) and six open-domain QA benchmarks.
• Findings: Search-o1 shows strong performance and improves the trustworthiness and applicability of LRMs in complex reasoning tasks.

Insights for Search / Rec / Personalization

• Architecture: cascaded retrieval + rerank + generation remains mainstream, but the agentic paradigm treats "when and how often to retrieve" as a learnable decision.
• Data: high-quality instruction data and session logs matter; synthetic data requires care against leakage and distribution shift.
• Evaluation: offline metrics increasingly diverge from online satisfaction; process-level metrics (answer success, citation accuracy, reasoning-chain completeness) are gaining importance.
• Deployment: latency, cost, interpretability, and safety are hard constraints for production systems, beyond benchmark optimization.

Limitations and Future Directions

Potential limitations include experiment scale, benchmark-versus-real-user distribution gaps, English-centric data limiting cross-lingual generalization, and safety risks when agents search the open web. Future work may explore more efficient test-time compute allocation, deeper integration with knowledge graphs and structured databases, and causal/fairness constraints for recommender applications.

Original Abstract (excerpt)

> 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... The code is available at https://github.com/sunnynexus/Search-o1.

Related Entries

• 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
• Retrieval Augmented Generation and Understanding in Vision: A Survey (arXiv:2503.18016)

Glossary

| Term | Meaning | |------|---------| | LRM | Large Reasoning Model | | RAG | Retrieval-Augmented Generation | | Agentic Search | Modeling search as sequential decision-making and tool calls | | Reason-in-Documents | Search-o1 module that analyzes retrieved documents before injecting them into the reasoning chain |