What is Fine-tuning?

Further training a base model on your own data so it specializes in a task, tone or domain.

Fine-tuning takes an already trained base model and continues training it on a smaller, task-specific dataset so its behavior shifts toward your needs. Instead of learning language from scratch, the model adjusts to your examples: a support tone, a document format, a classification scheme, or domain-specific phrasing. It typically needs hundreds to thousands of good example input-output pairs, and quality matters more than volume. Efficient methods like LoRA update only a small set of weights, making fine-tuning affordable on open-weight models; major API providers also offer hosted fine-tuning for some models. Fine-tuning is best for teaching consistent style, format, or behavior. It is a poor fit for injecting fresh or changing facts, where RAG usually works better. Many teams try prompting and RAG first, and fine-tune only when those fall short.

Example

A legal-tech startup fine-tunes a model on 2,000 examples of contract clauses labeled by risk level, so it flags clauses consistently without a long instruction prompt every time.

Why it matters

Fine-tuning can make a cheaper model perform like a bigger one on your specific task, but it adds cost and maintenance, so know when prompting or RAG is enough. Browse the AI tools directory or the model leaderboard to put it into practice.