What is Fine-Tuning?
Further training a base model on your own data to specialize it.
Fine-tuning takes a pre-trained model and trains it further on your own examples so it adapts to a specific style, domain, or task.
It's more involved than prompting but can improve consistency for narrow use cases.
Fine-tuning takes an already-trained model and trains it further on a focused set of examples, nudging its internal weights so it gets better at a specific task, style or domain. Rather than building a model from scratch — which costs enormous compute and data — you start from a capable foundation model and specialise it. Show it thousands of examples of your support tickets and ideal responses, and it learns to respond in that style; show it domain-specific text, and it picks up that domain's language.
Crucially, fine-tuning is best for teaching a model how to behave — tone, format, structure, a consistent way of responding — rather than for injecting new facts, which it tends to learn unreliably and which go stale. For up-to-date or specific factual knowledge, retrieval-augmented generation is usually the better tool. Fine-tuning has become more accessible and cheaper, and lightweight methods now let teams adapt models without the full cost of retraining everything.
Why it matters
Fine-tuning is one of the main ways organisations adapt general AI models to their specific needs, and knowing what it's good and bad at saves expensive mistakes. The common error is fine-tuning to add facts (which RAG does better) instead of to shape behaviour (which fine-tuning does well). It matters whenever you need an AI to consistently sound or act a certain way that prompting alone can't reliably achieve.
A concrete example
A company wants its AI to always reply to customers in a specific brand voice — warm, concise, never over-promising. Prompting helps, but inconsistently. By fine-tuning a model on thousands of real, on-brand replies, they bake that style into the model itself, so it produces on-voice responses reliably without a long instruction every time.
Where you’ll meet it in AI tools
Most users never fine-tune directly; they benefit from it inside products. Many specialised AI tools are fine-tuned versions of general models, and some platforms let businesses fine-tune a model on their own data. When a tool offers a distinct "voice" or domain expertise that general assistants lack, fine-tuning is often part of how it got there — alongside RAG and careful prompting.
Fine-Tuning: FAQ
Should I fine-tune or just use a better prompt?
Try prompting first — it's free, instant and often enough. Fine-tuning is worth it when you need a consistent behaviour, style or format that prompting can't reliably deliver, especially at scale where repeating a long instruction every time is impractical. If your need is up-to-date or specific facts rather than behaviour, neither prompting nor fine-tuning is ideal — use retrieval-augmented generation. In short: prompt for quick steering, fine-tune for consistent behaviour, RAG for knowledge.
Does fine-tuning teach a model new facts?
It can, but unreliably, and it's usually the wrong tool for facts. Fine-tuning excels at shaping how a model responds — tone, format, task behaviour — but facts learned this way can be absorbed imperfectly and go out of date, with no way to cite a source. For specific or current factual knowledge, retrieval-augmented generation, which feeds the model real documents at answer time, is more accurate and updatable. Reserve fine-tuning for behaviour and style.
Related terms
Find building with ai tools
Browse hand-reviewed AI tools — compare on pricing, features and real ratings.