AI summaryⓘ
The authors studied how well different language model retrieval tools work with company data, focusing on which ones can be legally used in businesses and how much they cost to build. They found that many top-performing systems rely on a non-commercially licensed tool without saying so, making them hard to use commercially. They also tested several embedders and found that until mid-2026, the best commercially licensed tool was noticeably less accurate than the usual free one, but a new tool released then matched the free one in quality and licensing. Finally, the authors highlight big differences in the costs of indexing data, noting that some systems hide these expenses, which can vary hugely when scaling up. They also differentiate between one-time indexing costs and ongoing answering costs, showing that the former is much more expensive at large scales.
language modelsretrieval systemsembeddingmulti-hop retrievalcommercial licensingRecall@5indexing costself-hostingAPI embeddersMuSiQue benchmark
Abstract
Enterprises connect language models to their own data through retrieval. The benchmarks that rank multi-hop retrieval systems leave out two facts a buyer needs before a published number can be used: whether the retrieval backbone may be deployed commercially, and what it costs to build. On licensing: the field's dense-retrieval anchor, NV-Embed-v2, is licensed cc-by-nc-4.0. Of the four leading MuSiQue systems we audit (HippoRAG-2, PropRAG, SAG, KET-RAG), three depend on it for their best numbers and none says so. On performance: we measure thirteen embedders from eight makers on one identical MuSiQue harness with bootstrap confidence intervals throughout. Until mid-2026 there was a real commercial tax: the best commercially-licensed embedder trailed the anchor by 2.31 Recall@5 points (95% CI [0.91, 3.71], p=0.001). NVIDIA's Nemotron-3-Embed-8B, released 2026-07-16, has closed it: +0.24 at Recall@5 (95% CI [-0.94, +1.43], p=0.69), -0.58 at Recall@10 (p=0.28). It matches the anchor, does not beat it, and is the only entrant that is commercially licensed, free to self-host, and indistinguishable from the anchor; every other entrant meeting the first two conditions sits 5.2 to 14.6 points below. The durable finding is the paid-versus-free divide: API embedders charge per token on every re-index, self-hosted ones charge nothing. On cost: three of five audited systems (adding Microsoft's GraphRAG) do not disclose indexing cost, and the only published GraphRAG dollar figures span 11x inside one third-party paper (USD 2.30 vs USD 24.94 to index a 5.64 MB corpus once); extrapolated to 1 TB that undisclosed choice separates roughly USD 428K from $4.6M. Our cost model keeps one-time embedding apart from recurring answering: at 1 TB, embedding sits 7.5x-900x below graph construction, and a year of answering at 10,000 queries/day sits 350x or more below it.