DecisionRAG replaces the retriever with a decision model, beating 8B rerankers by 13.8 R@1
omarsar0 · x · 2026-10-29
A new Cornell/ASAP paper, "Retriever-Free Retrieval: Navigating Documents with Decision Models," proposes DecisionRAG, which replaces the first-stage retriever with a pretrained decision model that navigates the corpus from broad topics down to documents and passages, scoring all candidates in a single forward pass without generating tokens.
Key results:
- With no task-specific training and no first-stage retriever, it outperforms dense retrieval plus an 8B reranker by 13.8 R@1 on MultiDoc2Dial and 17.4 Hit@1 on Natural Questions, searching 2.68M Wikipedia paragraphs.
- Retrieval takes just 0.7–1.8 seconds per query across four benchmarks.
The core insight: judging candidates jointly in shared context makes relevance estimates comparable, closing the capability asymmetry between retrievers and generators. Highly relevant for anyone building RAG systems.
Related event: DecisionRAG Replaces Retrievers with Decision Models(2 posts)→
More from coding & agent
- LukeW's next AI speaker series: 5 engineers on pushing agents past Crazy Agent Scale — LukeW · 2026-10-29
- AgentWeb launches: passes 47/47 live-site agent tasks at 1.9s median — damnGruz · 2026-10-29
- Vibe coding debate: understand the system, not the code, to vet AI fixes — brandon_galang · 2026-10-29
- Mintive launches: every AI agent gets its own identity and permissions, raises $11M — eptwts · 2026-10-29
- Harbor framework v0.25.0 ships 10 measures to curb reward hacking in agent evals — nirmal_dist · 2026-10-29
- LangChain's Managed Deep Agents can now control a browser for API-less sites — LangChain · 2026-10-29