{"paper":{"title":"Stop Overthinking: Unlocking Efficient Listwise Reranking with Minimal Reasoning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"A length-regularized self-distillation method lets student rerankers match teacher effectiveness while cutting reasoning tokens by 34-37%.","cross_cats":[],"primary_cat":"cs.IR","authors_text":"Danyang Liu, Kan Li","submitted_at":"2026-05-14T06:44:44Z","abstract_excerpt":"Listwise reranking utilizing Large Language Models (LLMs) has achieved state-of-the-art retrieval effectiveness. Recently, reasoning-enhanced models have further pushed these boundaries by employing Chain-of-Thought (CoT) to perform deep comparative analysis of candidate documents. However, this performance gain comes at a prohibitive computational cost, as models often generate thousands of reasoning tokens before producing a final ranking. In this work, we investigate the relationship between reasoning length and ranking quality, revealing an overthinking phenomenon where extended reasoning "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Experiments on TREC Deep Learning and NeuCLIR benchmarks demonstrate that our method maintains the teacher's effectiveness while reducing inference token consumption by 34%-37% across different retrieval settings.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The Pareto-inspired filter reliably selects reasoning traces that preserve ranking quality when transferred to the student model; this transfer is assumed to work without additional validation on held-out data or different model scales.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Length-regularized self-distillation trains rerankers to use concise reasoning traces that match teacher performance at 34-37 percent lower token cost.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A length-regularized self-distillation method lets student rerankers match teacher effectiveness while cutting reasoning tokens by 34-37%.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"16c3e37d41985ab85009f9bd3b8ddd657f61dacff6c179a940328c509d76cacb"},"source":{"id":"2605.14450","kind":"arxiv","version":1},"verdict":{"id":"702478b1-86a5-4569-908c-94ecf807b62a","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T02:07:57.820930Z","strongest_claim":"Experiments on TREC Deep Learning and NeuCLIR benchmarks demonstrate that our method maintains the teacher's effectiveness while reducing inference token consumption by 34%-37% across different retrieval settings.","one_line_summary":"Length-regularized self-distillation trains rerankers to use concise reasoning traces that match teacher performance at 34-37 percent lower token cost.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The Pareto-inspired filter reliably selects reasoning traces that preserve ranking quality when transferred to the student model; this transfer is assumed to work without additional validation on held-out data or different model scales.","pith_extraction_headline":"A length-regularized self-distillation method lets student rerankers match teacher effectiveness while cutting reasoning tokens by 34-37%."},"references":{"count":2,"sample":[{"doi":"","year":2019,"title":"R. Nogueira and K. Cho, ‘Passage Re-ranking with BERT’, ArXiv Prepr. ArXiv190104085, 2019. [19] S. Hongjin et al., ‘BRIGHT: A Realistic and Challenging Benchmark for Reasoning-Intensive Retrieval’, in","work_id":"2ae19ad0-6558-413f-9850-4c9ee7dfcee8","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Chen et al., ‘TourRank: Utilizing Large Language Models for Documents Ranking with a Tournament-Inspired Strategy’, ArXiv Prepr","work_id":"dc4751f9-ee5d-40d6-8bee-e77c503fa779","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":2,"snapshot_sha256":"dad3146a84a35bfb7c8ce7a2f36dccfb4afb515df42c6a50b4642025944dec83","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}