When a shorter fine-tuned prompt beats a long few-shot prompt.
The real economic case for fine-tuning is deleting a giant few-shot prompt you'd otherwise resend forever. Do the token math first — see the fine-tuning cost calculator.
Free prompt-vs-fine-tune calculator — compare a long few-shot prompt against fine-tuning to find the break-even volume.
When it lets you drop a long few-shot prompt for a short one. You pay a one-off training cost, then save on input tokens every call. At high volume the payback is fast; at low volume the long prompt is fine.
The training is a one-time cost, and fine-tuned models sometimes charge a bit more per token. The saving comes from a shorter prompt — so it wins when prompt length × volume is large.