Hybrid Deep Searcher: Scalable Parallel and Sequential Search Reasoning
#HybridDeepSearcher #Retrieval-augmented generation #Large reasoning models #Parallel search #Sequential reasoning #HDS-QA #ICLR 2026 #Test-time scaling
📌 Key Takeaways
- HybridDeepSearcher integrates parallel query expansion with evidence aggregation
- Researchers introduced HDS-QA dataset for supervised reasoning
- Approach significantly outperforms state-of-the-art on multiple benchmarks
- Demonstrates consistent test-time search scaling unlike competing methods
- Paper accepted to ICLR 2026 conference
📖 Full Retelling
Researchers led by Dayoon Ko and eight collaborators have developed HybridDeepSearcher, a novel hybrid search strategy for artificial intelligence reasoning models, addressing limitations in existing retrieval-augmented generation approaches, as detailed in their paper submitted to arXiv on August 26, 2025 and revised on February 24, 2026. The research team from various institutions identified that current methods for extending reasoning through single-query sequential search suffer from limited evidence coverage, while approaches generating multiple independent queries per step often lack structured aggregation, hindering deeper sequential reasoning capabilities in large reasoning models. To overcome these challenges, the researchers proposed a hybrid search strategy that combines the strengths of both parallel and sequential search approaches. Their solution introduces HybridDeepSearcher, a structured search agent that integrates parallel query expansion with explicit evidence aggregation before advancing to deeper sequential reasoning, creating a more comprehensive and effective knowledge retrieval process. To train and evaluate this new approach, the researchers developed HDS-QA, a novel dataset that guides models to combine broad parallel search with structured aggregation through supervised reasoning-query-retrieval trajectories containing parallel sub-queries. When tested across five benchmarks, HybridDeepSearcher significantly outperformed existing state-of-the-art methods, improving F1 scores by +15.9 on FanOutQA and +9.2 on a subset of BrowseComp. Notably, the analysis revealed consistent test-time search scaling, where performance continued to improve as additional search turns or calls were allowed, while competing methods plateaued early, demonstrating the superior scalability and adaptability of the new approach.
🏷️ Themes
Artificial Intelligence, Machine Learning, Knowledge Retrieval, Research Methodology
📚 Related People & Topics
Reasoning model
Language models designed for reasoning tasks
A reasoning model, also known as reasoning language models (RLMs) or large reasoning models (LRMs), is a type of large language model (LLM) that has been specifically trained to solve complex tasks requiring multiple steps of logical reasoning. These models demonstrate superior performance on logic,...
Entity Intersection Graph
Connections for Reasoning model:
🌐
Reinforcement learning
2 shared
Original Source
--> Computer Science > Artificial Intelligence arXiv:2508.19113 [Submitted on 26 Aug 2025 ( v1 ), last revised 24 Feb 2026 (this version, v2)] Title: Hybrid Deep Searcher: Scalable Parallel and Sequential Search Reasoning Authors: Dayoon Ko , Jihyuk Kim , Haeju Park , Sohyeon Kim , Dahyun Lee , Yongrae Jo , Gunhee Kim , Moontae Lee , Kyungjae Lee View a PDF of the paper titled Hybrid Deep Searcher: Scalable Parallel and Sequential Search Reasoning, by Dayoon Ko and 8 other authors View PDF Abstract: Large reasoning models combined with retrieval-augmented generation have enabled deep research agents capable of multi-step reasoning with external knowledge retrieval. However, we find that existing approaches rarely demonstrate test-time search scaling. Methods that extend reasoning through single-query sequential search suffer from limited evidence coverage, while approaches that generate multiple independent queries per step often lack structured aggregation, hindering deeper sequential reasoning. We propose a hybrid search strategy to address these limitations. We introduce HybridDeepSearcher, a structured search agent that integrates parallel query expansion with explicit evidence aggregation before advancing to deeper sequential reasoning. To supervise this behavior, we introduce HDS-QA, a novel dataset that guides models to combine broad parallel search with structured aggregation through supervised reasoning-query0retrieval trajectories containing parallel sub-queries. Across five benchmarks, HybridDeepSearcher significantly outperforms the state-of-the-art, improving F1 scores by +15.9 on FanOutQA and +9.2 on a subset of BrowseComp. Further analysis shows its consistent test-time search scaling: performance improves as additional search turns or calls are allowed, while competing methods plateau. Comments: Accepted to ICLR 2026 Subjects: Artificial Intelligence (cs.AI) Cite as: arXiv:2508.19113 [cs.AI] (or arXiv:2508.19113v2 [cs.AI] for this version) https://do...
Read full article at source