Continuing training on your own examples to specialize a base model's style, format or domain — distinct from just writing a better prompt.
Fine-tuning updates a model's weights on a curated dataset of your examples, baking in a behavior rather than describing it in the prompt every time. It's the tool for locking in a house style, a rigid output format, or a narrow domain vocabulary that prompting alone keeps drifting away from.
It's usually the wrong first move. Prompting, few-shot examples and RAG solve most problems faster and cheaper, and a fine-tuned model is frozen — it won't benefit from the next base-model upgrade without redoing the work. Reach for it only after the cheaper levers plateau.
Fine-tuning needs open weights or a provider's tuning API. Open-weight models (on Hugging Face) give you full control; closed models offer it only where the maker exposes it.