RAG vs Fine-tuning

Retrieval-augmented generation vs fine-tuning an LLM.

RAG gives models access to new facts. Fine-tuning changes model behavior/style/format. They solve different problems.

At a glance

	RAG	Fine-tuning
Changes	What facts the model can access	How the model behaves
Cost	Low (infrastructure)	High (training)
Update frequency	Anytime (reindex)	Rare (retrain)
Use case	Facts, docs, grounding	Style, format, specialized tasks

You need the model to answer from specific (changing) facts.

You need stable output format/style the model can't get via prompting.

Start with RAG. Fine-tune only when RAG isn't enough and the behavior you need is stable.