RAG

Which LLM wins at RAG Q&A?

Kari Jaaskelainen

— 1 min read
Which LLM wins at RAG Q&A?

Which LLM wins at RAG Q&A?

RAG (Retrieval-Augmented Generation) helps reduce hallucinations by grounding answers in source documents. This study tested five 7B-class models on computer science literature Q&A to compare accuracy and speed.

  • GPT‑3.5 + RAG answered both yes/no and long-form questions effectively.
  • Mistral‑7B‑Instruct + RAG led the open-source pack on both question types.
  • Orca‑mini‑v3‑7B was fastest (lowest average latency); LLaMa2‑7B‑Chat was slowest.

How they measured it: accuracy and precision for binary questions; human expert and Gemini rankings; and cosine similarity for long answers.

Big picture: With the right RAG setup and infrastructure, open-source LLMs can stand shoulder to shoulder with proprietary models like GPT‑3.5.

Paper by Ranul Dayarathne, Uvini Ranaweera, and Upeksha Ganegoda. Read more: http://arxiv.org/abs/2511.03261v1

Paper: http://arxiv.org/abs/2511.03261v1

Register: https://www.AiFeta.com

RAG LLMs QA GenerativeAI Mistral GPT35 OpenSourceAI AIResearch NLP ComputerScience

Read more