Fine-tuning or prompting? A practical decision guide
Guides5 mins read
Introduction
Most teams reach for prompting first, and they should. It's fast, cheap to change and easy to review. But prompts hit a ceiling when a task needs a consistent format, specialised vocabulary or behaviour that's hard to describe in words.
Signals you've outgrown prompting
Your prompt has grown past a page and still misses edge cases. Outputs vary run to run on identical inputs. You're paying for long contexts just to restate the same instructions.
If two of those are true, a small supervised fine-tune on a few hundred good examples usually beats another round of prompt edits.
A simple test
Measure before you train
Build an evaluation set of 50–100 real cases with known good answers. Score your best prompt against it, then score a fine-tuned model. Decide on the numbers, not the anecdotes.
