OpenRouter tested RAG re-ranking: precision@1 rose from 81% to 97% on a benchmark of 274 questions.
Automated summary · Based on collected summaries; not individually verified
Prepared by HotAI
What happened
OpenRouter has released a tutorial on RAG re-ranking. On a benchmark of 274 questions, vector retrieval using google/gemini-embedding-2 achieved a precision@1 of 81.0%. Adding voyageai/rerank-3-lite to re-rank the top 30 results improved this to 97.1%, with costs around $0.10 per thousand queries.
Background
What is an embedding? Vector representations for similarity and retrieval
Embeddings map content into numerical vectors for learned similarity comparisons, often supporting retrieval or clustering. Similarity is not factual equivalence or correctness. Use compatible models and metrics rather than mixing equal-dimensional vectors from different models. Keep source text separately; embeddings are not losslessly recoverable document archives.
What is precision? How many selected results are truly positive
Precision is TP/(TP+FP): true positives among predicted positives. It differs from overall accuracy and recall. Define the positive class and references. Fewer false positives can improve precision, but missed targets still matter; one high ratio is not complete system quality.
Reading notes
The event summary comes from collected excerpts. Background passages quote existing HotAI articles to explain terms and do not establish new facts about this event. Check the original sources for product results, conditions and subsequent updates.
Original source
Sources and collection timeline
| Source | Report title | Original publication time | Collected by HotAI |
|---|---|---|---|
| openrouter.ai | OpenRouter tested RAG re-ranking: precision@1 rose from 81% to 97% on a benchmark of 274 questions. | 2026-11-13 08:00 | 2026-11-13 09:07 |