What is Fine-tuning?
Continuing the training of a base LLM on your own labeled examples to specialize its behavior — different from prompt engineering (changing instructions) or RAG (providing context at inference time). Useful when you need a specific tone, format, or domain vocabulary the base model can't reliably hit via prompting alone. Available through the OpenAI API and some open-weight models; not available in ChatGPT or Claude Cowork. Costs 100×-1000× a regular inference call but is a one-time cost.
What it is
Fine-tuning is further training of a pretrained language model on additional data, which makes the model start to represent and imitate the patterns of that data. Anthropic notes that Claude is not a bare language model: it has already been fine-tuned to be a helpful assistant, and RLHF is used on top of pretraining for the same reason.
When it helps
It can adapt a model to a specific domain, task or writing style, but it needs careful thought about the training data and the effect on the model's performance and biases. For most agent work, better instructions, examples in the prompt, or retrieval from your own documents get you most of the way with far less effort.
Availability varies
Not every provider offers it. At the time of writing, Anthropic's glossary states that the Claude API does not currently offer fine-tuning and invites customers to ask their Anthropic contact. Check your provider's documentation, and treat fine-tuning as a different thing from an agent's memory, which stores notes in files or a database and leaves the model's weights untouched.
Example
A team wants every answer in their house style. Before fine-tuning anything, they try a system prompt with three example answers; if that is not enough, they check whether their model provider offers fine-tuning at all.
Sources
See also
← Back to the full AI agent glossary.