Retriever-free retrieval: a paper swaps the RAG retriever for a decision model that walks the corpus
publicado (OK do Fabricio 30/10 BRT).
“Retriever-Free Retrieval: Navigating Documents with Decision Models” (Chao Wan, Paul Jünger and others) went up on arXiv on 26 October 2026 and got wider attention on 29 Oct when Elvis Saravia shared it (about 390 likes).
The idea. Instead of a first-stage retriever, a decision model walks the corpus from broad topics down to documents and then passages. At each level it scores every candidate in one forward pass, without generating tokens, and keeps only the most promising branches.
Authors’ numbers (UnverifiedClaim). With no first-stage retriever and no task-specific training, it beats dense retrieval plus an 8B reranker by 13.8 R@1 points on MultiDoc2Dial and by 17.4 Hit@1 points on Natural Questions, searching 2.68 million Wikipedia paragraphs. Retrieval takes 0.7 to 1.8 seconds and 0.2 to 2.3 cents per query. Scoring candidates together in one shared context adds up to 7.7 points over judging them one at a time. A Qwen3-8B judge trained on in-domain data reaches 60.7 R@1 on MultiDoc2Dial.
Why it matters here. It is a use pattern, not a model: decision models as navigators rather than rerankers. Compare the earlier paper arguing decision models are not better rerankers on their own.