{"paper":{"title":"Scaling Retrieval-Augmented Reasoning with Parallel Search and Explicit Merging","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"MultiSearch retrieves external knowledge from multiple query perspectives in parallel and merges the results to raise signal-to-noise ratio before reasoning.","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Chunxu Shen, Jiabei Liu, Jiancan Wu, Junfei Tan, Lingling Yi, Wenyu Mao, Xiang Wang","submitted_at":"2026-05-13T13:46:35Z","abstract_excerpt":"Deep search agents have proven effective in enhancing LLMs by retrieving external knowledge during multi-step reasoning. However, existing methods often generate a single query for retrieval at each reasoning step, limiting information coverage and introducing high noise. This may result in low signal-to-noise ratios (SNR) during search, degrading reasoning accuracy and leading to unnecessary reasoning steps. In this paper, we introduce MultiSearch, an RL-based framework that addresses these limitations through multi-query retrieval and explicit merging of retrieved information. At each reason"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"MultiSearch outperforms baseline methods, enhancing the SNR of retrieval and improving reasoning performance in question-answering tasks.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That generating multiple queries from different perspectives and then performing explicit merging will consistently raise signal-to-noise ratio without introducing new noise sources or prohibitive compute overhead in the merging step.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"MultiSearch uses parallel multi-query retrieval plus explicit merging inside a reinforcement-learning loop to improve retrieval-augmented reasoning, outperforming baselines on seven QA benchmarks.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"MultiSearch retrieves external knowledge from multiple query perspectives in parallel and merges the results to raise signal-to-noise ratio before reasoning.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"4447f975051f44b691ac6509fb9d35f265808db1e87e7a8011424fe6c8bd4778"},"source":{"id":"2605.13534","kind":"arxiv","version":1},"verdict":{"id":"cfb4a029-898a-4fcd-b535-0ff45b2448f8","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T18:49:24.839275Z","strongest_claim":"MultiSearch outperforms baseline methods, enhancing the SNR of retrieval and improving reasoning performance in question-answering tasks.","one_line_summary":"MultiSearch uses parallel multi-query retrieval plus explicit merging inside a reinforcement-learning loop to improve retrieval-augmented reasoning, outperforming baselines on seven QA benchmarks.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That generating multiple queries from different perspectives and then performing explicit merging will consistently raise signal-to-noise ratio without introducing new noise sources or prohibitive compute overhead in the merging step.","pith_extraction_headline":"MultiSearch retrieves external knowledge from multiple query perspectives in parallel and merges the results to raise signal-to-noise ratio before reasoning."},"references":{"count":47,"sample":[{"doi":"","year":2024,"title":"A survey on evaluation of large language models.ACM transactions on intelligent systems and technology, 15(3):1–45","work_id":"a9075135-e1ef-4f62-89ff-20df37c58319","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Siren’s song in the ai ocean: A survey on hallucination in large language models.Computational Linguistics, 51(4):1373–1418, 2025","work_id":"0ddc813d-938a-47c5-b4a0-693d9a296cfd","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"A study of generative large language model for medical research and healthcare.NPJ digital medicine, 6(1):210, 2023","work_id":"06400e59-275e-40c4-9fa0-50ae5f9f4128","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Retrieval-Augmented Generation for Large Language Models: A Survey","work_id":"b80d2790-6cd9-4c87-b3c4-de404f99a80e","ref_index":4,"cited_arxiv_id":"2312.10997","is_internal_anchor":true},{"doi":"","year":2020,"title":"Retrieval-augmented generation for knowledge-intensive nlp tasks.Advances in neural information processing systems, 33:9459–9474","work_id":"f4e7687c-6c0e-4b2d-92f6-01888410340c","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":47,"snapshot_sha256":"09bd96588df310aad790ab67f530742764c40f0936eb1c84757f7770a33c50dc","internal_anchors":13},"formal_canon":{"evidence_count":2,"snapshot_sha256":"1d77c544fa3c3b297707671fad1749160766881d74a58c621c4146a3e839d2ef"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}