Compare · Building
RAG vs fine-tuning
Both techniques customize how a model behaves — but they solve different problems. Use this page to choose a starting point, then practice with interactive lessons.
Quick comparison
- RAG — retrieve relevant docs at answer time. Strong for factual, changing knowledge and citations.
- Fine-tuning — update model weights on examples. Strong for style, format, and repeated task behavior.
- Cost — RAG needs retrieval infra; fine-tuning needs training runs and ongoing evals.
- Freshness — RAG updates when you reindex; fine-tunes go stale until you retrain.
- Default — prompt well first; add RAG for changing facts; fine-tune when format still drifts. See also prompting vs RAG vs fine-tuning.
When RAG wins
Policies, product docs, tickets, and research corpora change weekly. If the model must ground answers in those sources, start with What is RAG? and the how-to build a 5-doc RAG.
When fine-tuning wins
You need consistent JSON shapes, tone, or domain shorthand that prompting cannot lock in. Learn the tradeoffs in Fine-tuning vs RAG and the how-to train a LoRA adapter.
FAQ
Should I use RAG or fine-tuning?
Start with RAG when answers must cite changing documents. Consider fine-tuning when you need a stable style, format, or domain behavior that prompting alone cannot hold — and you can afford training and evals.
Can I combine RAG and fine-tuning?
Yes. Fine-tune for behavior and style; retrieve for facts. Many production systems use both, with evaluations on faithfulness and format.