FINE-TUNING

Extra training on your own examples that adjusts a model weights so it handles a specific task or style better.

Zero retention: everything you type stays in this browser. Verify it yourself in DevTools.
Plain-language definition

What Does Fine-Tuning Mean?

Fine-tuning means taking a model that has already been trained and training it a bit more on your own examples, so it gets better at a particular task, format or style. Unlike a prompt or an instructions file, which the model reads each time, fine-tuning changes the model itself, specifically its weights, the numbers that determine its behavior. You typically need hundreds to thousands of good input and output pairs. Some providers offer fine-tuning through an API for selected models, and open-weight models can be fine-tuned on your own hardware with techniques such as LoRA that adjust only a small part of the model.

Why Does Fine-Tuning Matter?

Fine-tuning is often reached for too early. Many problems people expect to need it for are solved more cheaply with better instructions, a few examples in the prompt, or retrieval of the right documents. Fine-tuning shines for consistent formats, narrow classification and cutting costs by making a small model good at one job. It is a poor way to teach a model facts that change, since you would have to retrain to update them.

What Does Fine-Tuning Look Like in Practice?

A company wants support replies in a very specific tone and structure. Prompting gets them 80 percent of the way, but outputs drift. They collect 2,000 approved past replies, fine-tune a smaller model on them, and now get the format reliably at a lower cost per reply than prompting a large model with long instructions. They keep the original prompt-based setup as a fallback and compare both on an eval each time they retrain.

What Do People Get Wrong About Fine-Tuning?

Fine-tuning is not the best way to give a model your documentation or codebase knowledge. Facts learned this way are unreliable and go stale. Retrieval or an instructions file is usually the better tool for knowledge.

Which Terms Are Related?

  • Distillation: Training a small model to imitate a larger model, so you get much of the quality at a fraction of the cost.
  • RAG (retrieval-augmented generation): Looking up relevant documents first and giving them to the model, so its answer is grounded in real sources.
  • Open-weight model: A model whose trained weights you can download and run on your own hardware, under the terms of its license.
  • System prompt: The standing instructions an application gives a model before the conversation, setting its role, rules and format.
  • Context engineering: Deciding what goes into a model context window, including instructions, files and tool output, so the model can do the task well.

All 28 terms are in the AI glossary. For models side by side, see frontier vs open-weight.

Frequently Asked Questions

Do I need fine-tuning to make a coding agent follow my rules?

Almost never. Put your commands and rules in an instructions file such as AGENTS.md or CLAUDE.md. It is free, takes minutes, and can be changed any time, while fine-tuning is slow and costly.

How much data does fine-tuning need?

It depends on the task, but useful results often start at a few hundred high-quality examples, and more varied tasks need thousands. Quality and consistency of examples matter more than raw volume.

Can fine-tuning make a model worse?

Yes. Training on narrow data can reduce general ability or safety behavior, and poor examples teach poor habits. Always compare the tuned model against the original on an eval before switching.

Was this helpful?