REVIEW 20 cited by
ReasonIR: Training Retrievers for Reasoning Tasks
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
We present ReasonIR-8B, the first retriever specifically trained for general reasoning tasks. Existing retrievers have shown limited gains on reasoning tasks, in part because existing training datasets focus on short factual queries tied to documents that straightforwardly answer them. We develop a synthetic data generation pipeline that, for each document, our pipeline creates a challenging and relevant query, along with a plausibly related but ultimately unhelpful hard negative. By training on a mixture of our synthetic data and existing public data, ReasonIR-8B achieves a new state-of-the-art of 29.9 nDCG@10 without reranker and 36.9 nDCG@10 with reranker on BRIGHT, a widely-used reasoning-intensive information retrieval (IR) benchmark. When applied to RAG tasks, ReasonIR-8B improves MMLU and GPQA performance by 6.4% and 22.6% respectively, relative to the closed-book baseline, outperforming other retrievers and search engines. In addition, ReasonIR-8B uses test-time compute more effectively: on BRIGHT, its performance consistently increases with longer and more information-rich rewritten queries; it continues to outperform other retrievers when combined with an LLM reranker. Our training recipe is general and can be easily extended to future LLMs; to this end, we open-source our code, data, and model.
Forward citations
Cited by 20 Pith papers
-
Frustratingly Simple Retrieval Improves Challenging, Reasoning-Intensive Benchmarks
A compact, diverse, web-scale datastore plus simple dense retrieval improves performance on MMLU, MMLU Pro, AGI Eval, GPQA, and MATH across multiple LLM families.
-
CodeGrep: An RL-Trained Retrieval Agent for LLM Coding Agents
An RL-trained 14B file-retrieval agent injected into a frozen coding agent reduces resolved-instance token use by about 19 percent on SWE-Bench Verified with no measured resolve-rate loss.
-
Antares: Foundation Models for Agentic Vulnerability Localization
Antares-3B, a 3B model trained with SFT plus GRPO, matches GPT-5.5 on repository-scale vulnerability localization at roughly 1/100th the inference cost.
-
TabRank: Chain-of-Thought Distillation for Table Re-Rankers
Conditioning a table reranker on teacher chain-of-thought traces during training, without supervising generation of the traces, improves out-of-distribution table retrieval accuracy over base, naive SFT, and standard ...
-
Procedural Knowledge at Scale Improves Reasoning
Retrieving compact procedural hints from 32M subquestion–subroutine pairs improves reasoning-model accuracy on math, science, and coding benchmarks beyond compute-matched test-time scaling.
-
Revisiting Text Ranking in Deep Research
In deep-research agents, passage-level BM25 retrieval plus re-ranking beats larger neural rankers, and converting agent queries into natural questions (Q2Q) improves neural ranking.
-
With Argus Eyes: Assessing Retrieval Gaps via Uncertainty Scoring to Detect and Remedy Retrieval Blind Spots
An entity-level Retrieval Probability Score exposes systematic blind spots in dense retrievers, predicts them from embedding geometry, and targeted Wikipedia-context augmentation of flagged entities raises nDCG@5/10 o...
-
Benchmarking Information Retrieval Models on Complex Retrieval Tasks
CRUMB is a new benchmark for complex, multi-aspect retrieval tasks on which state-of-the-art retrieval models score poorly, and query rewriting does not rescue the best models.
-
ERank: Fusing Supervised Fine-Tuning and Reinforcement Learning for Effective and Efficient Text Reranking
ERank trains a pointwise reasoning-LLM reranker with fine-grained integer scoring and a listwise reinforcement-learning reward, reporting nDCG@10 of 40.2 on BRIGHT for its 32B variant with BM25 hybrid scoring.
-
BrowseComp-Plus: A More Fair and Transparent Evaluation Benchmark of Deep-Research Agent
BrowseComp-Plus is a fixed, human-verified document corpus with supporting and negative documents that allows controlled, reproducible evaluation of deep-research agents and their retrievers.
-
TongSearch-QR: Reinforced Query Reasoning for Retrieval
TongSearch-QR trains 1.5B and 7B models with GRPO and a frozen-embedding reward to rewrite queries, reaching 27.9 nDCG@10 on BRIGHT with BM25, above GPT-4o's 26.5.
-
From Token to Action: State Machine Reasoning to Mitigate Overthinking in Information Retrieval
A state-machine framework that replaces token-level chain-of-thought with discrete query-refinement and reranking actions reduces token use by 74% while improving nDCG@10 on retrieval benchmarks.
-
RaDeR: Reasoning-aware Dense Retrieval Models
A math-trained dense retriever and reranker, built from MCTS reasoning trajectories and self-reflection, outperforms strong baselines on reasoning-intensive retrieval benchmarks and beats BM25 on chain-of-thought queries.
-
Don't "Overthink" Passage Reranking: Is Reasoning Truly Necessary?
Standard pointwise rerankers beat reasoning-based pointwise rerankers under identical training, and removing the reasoning process at inference partially recovers the gap.
-
Diffusion vs. Autoregressive Language Models: A Text Embedding Perspective
Diffusion-language-model embeddings beat LLM embeddings on long-document and reasoning retrieval benchmarks, but the paper overstates some gains and releases no code or data.
-
Rank-K: Test-Time Reasoning for Listwise Reranking
Rank-K, a reasoning-model-based listwise reranker distilled from DeepSeek R1 traces, beats RankZephyr on several benchmarks but only marginally on TREC DL 2019/2020.
-
SHIFT: Self-reconstruction Harnesses Implicit Fine-grained Thinking for Retrieval
SHIFT trains LLM retrievers to reason through latent continuous tokens and reconstruct explicit reasoning traces, improving reasoning-intensive retrieval.
-
Beyond Sequential Reranking: Reranker-Guided Search Improves Reasoning Intensive Retrieval
Reranker-Guided-Search, a greedy graph search steered by reranker scores, outperforms sequential top-k reranking under a fixed budget on three reasoning-intensive retrieval benchmarks.
-
Domain-Aware RAG: MoL-Enhanced RL for Efficient Training and Scalable Retrieval
A two-stage RAG training pipeline, MoL continual pre-training plus GRPO reinforcement learning with single-passage training and multi-passage inference, reports state-of-the-art retrieval recall on NFCORPUS and SCIFAC...
-
A Vision for Geo-Temporal Deep Research Systems: Towards Comprehensive, Transparent, and Reproducible Geo-Temporal Information Synthesis
A research agenda calling for geo-temporal reasoning in deep research systems, with no experiments or system implementation.
Discussion (0). Sign in to comment.