{"id":5006,"date":"2026-11-19T18:04:50","date_gmt":"2026-11-19T18:04:50","guid":{"rendered":"https:\/\/blogs.ua.es\/faraonllorens\/?p=5006"},"modified":"2026-11-19T18:04:51","modified_gmt":"2026-11-19T18:04:51","slug":"when-is-llm-based-reasoning-retrieval-worth-the-cost-lightweight-cost-aware-query-routing-for-retrieval-augmented-generation","status":"publish","type":"post","link":"https:\/\/blogs.ua.es\/faraonllorens\/2026\/11\/19\/when-is-llm-based-reasoning-retrieval-worth-the-cost-lightweight-cost-aware-query-routing-for-retrieval-augmented-generation\/","title":{"rendered":"When is LLM-based reasoning retrieval worth the cost? Lightweight cost-aware query routing for retrieval-augmented generation"},"content":{"rendered":"\n<p class=\"wp-block-paragraph\"><strong>When is LLM-based reasoning retrieval worth the cost? Lightweight cost-aware query routing for retrieval-augmented generation<\/strong><br>Sergio Arjona-Giner, Fara\u00f3n Llorens-Largo, Alberto Real-Fern\u00e1ndez y Rafael Molina-Carmona (Universidad de Alicante)<br><em><a href=\"https:\/\/www.sciencedirect.com\/journal\/information-processing-and-management\" data-type=\"link\" data-id=\"https:\/\/www.sciencedirect.com\/journal\/information-processing-and-management\" target=\"_blank\" rel=\"noopener\">Information Processing &amp; Management<\/a>,<\/em> Volume 64, Issue 2, Part C, February 2027<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><a href=\"https:\/\/www.sciencedirect.com\/science\/article\/pii\/S0306457326006771\" target=\"_blank\" rel=\"noopener\">https:\/\/www.sciencedirect.com\/science\/article\/pii\/S0306457326006771<\/a><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Retrieval-Augmented Generation (RAG) systems face a cost\u2013effectiveness trade-off: inexpensive retrievers are efficient, whereas reasoning-based retrievers can improve retrieval quality at substantially higher computational cost. We introduce CARE (Cost-aware Adaptive REtrieval), a lightweight cost-aware binary routing method that decides whether each query should be handled by an inexpensive Hybrid retriever or escalated to the ReSP (Retrieve, Summarize, Plan) reasoning-based retriever. The routing decision is made by a lightweight logistic-regression gate using query-level and inexpensive probe-retrieval features. The evaluation covers two BEIR question-answering (QA) datasets, HotpotQA and Natural Questions (NQ), comprising 7,914,797 documents and 10,857 benchmark queries in total, of which 4097 constitute the fully held-out test sets used for evaluation. Using dataset-specific thresholds selected independently on the corresponding validation splits (\u03c4=0.30 for HotpotQA and \u03c4=0.60 for NQ), CARE achieves utilities of 0.8390 and 0.7596, respectively. These dataset-specific operating points yield a macro-average utility of 0.7993 and a macro-average expensive-branch activation rate of 49.31%. On HotpotQA, CARE approximately matches ReSP in retrieval effectiveness, although its Recall@100 is slightly lower (0.8770 versus 0.8795). Its 0.0007 utility advantage is therefore driven by the routing-cost term rather than by superior retrieval effectiveness, with expensive-branch activation reduced from 100% to 88.79%. On NQ, CARE provides a clearer cost-sensitive benefit, achieving 0.7596 utility with only 9.82% activation.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Keywords<\/strong><br>Retrieval-augmented generation, Adaptive retrieval, Query routing, Cost-aware retrieval, Retrieval efficiency<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/ars.els-cdn.com\/content\/image\/1-s2.0-S0306457326006771-gr1.jpg\" alt=\"\" \/><\/figure>\n","protected":false},"excerpt":{"rendered":"<p>When is LLM-based reasoning retrieval worth the cost? Lightweight cost-aware query routing for retrieval-augmented generationSergio Arjona-Giner, Fara\u00f3n Llorens-Largo, Alberto Real-Fern\u00e1ndez y Rafael Molina-Carmona (Universidad de Alicante)Information Processing &amp; Management, Volume 64, Issue 2, Part C, February 2027 https:\/\/www.sciencedirect.com\/science\/article\/pii\/S0306457326006771 Retrieval-Augmented Generation (RAG) systems face a cost\u2013effectiveness trade-off: inexpensive retrievers are efficient, whereas reasoning-based retrievers can improve [&hellip;]<\/p>\n","protected":false},"author":30,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[163438,2325692],"tags":[],"class_list":["post-5006","post","type-post","status-publish","format-standard","hentry","category-inteligencia-artificial","category-smart_learning"],"_links":{"self":[{"href":"https:\/\/blogs.ua.es\/faraonllorens\/wp-json\/wp\/v2\/posts\/5006","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/blogs.ua.es\/faraonllorens\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/blogs.ua.es\/faraonllorens\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/blogs.ua.es\/faraonllorens\/wp-json\/wp\/v2\/users\/30"}],"replies":[{"embeddable":true,"href":"https:\/\/blogs.ua.es\/faraonllorens\/wp-json\/wp\/v2\/comments?post=5006"}],"version-history":[{"count":1,"href":"https:\/\/blogs.ua.es\/faraonllorens\/wp-json\/wp\/v2\/posts\/5006\/revisions"}],"predecessor-version":[{"id":5007,"href":"https:\/\/blogs.ua.es\/faraonllorens\/wp-json\/wp\/v2\/posts\/5006\/revisions\/5007"}],"wp:attachment":[{"href":"https:\/\/blogs.ua.es\/faraonllorens\/wp-json\/wp\/v2\/media?parent=5006"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/blogs.ua.es\/faraonllorens\/wp-json\/wp\/v2\/categories?post=5006"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/blogs.ua.es\/faraonllorens\/wp-json\/wp\/v2\/tags?post=5006"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}