Elegance AI
Home/Lab/Fine-Tuning Decision Lab

When does fine-tuning actually earn its keep?

Same conversation, three models. Type your own or pick an example, then run all three live and watch what fine-tuning buys, and where it stops mattering.

Showing a saved example. Click Summarize to run it live through all three models.
Base modelOFF THE SHELF
facebook/bart-large-cnn

Lenny asks Bob to help him pick a new pair of trousers. Lenny already has two pairs of purple trousers. He also has four black pairs. Bob: What matters is what you'll give you most outfit options.

Cost / 1K
~negligible (compute only)
Latency
22250 ms
Size
400M
Hosting
self-hosted
Fine-tunedTRAINED
tuned on SAMSum dialogue

Lenny asks Bob to help him pick a pair of purple trousers. Bob says he has four black pairs of the same colour.

Cost / 1K
~negligible (compute only)
Latency
17681 ms
Size
400M
Hosting
self-hosted
Frontier modelZERO TRAINING
Claude Haiku · plain prompt

Lenny asks Bob for help choosing between three pairs of trousers by sending photos. Bob initially recommends the first pair, but after Lenny expresses concern about already owning purple trousers, they discuss whether to pick a different color for better outfit variety. Ultimately, Bob advises Lenny to choose based on quality, and Lenny decides to pick between the first or third pair.

Cost / 1K
~$0.72 / 1K summaries
Latency
5631 ms
Size
frontier
Hosting
API

The two BART columns are self-hosted on a free autoscaling service, so they are slow on CPU and cold-start when idle. Haiku runs through the API. The Inference Cost Modeller shows where that cost difference actually bites.

Read the bottom row, not the summaries. Fine-tuning clearly lifted the base model into a usable dialogue summary, for almost no cost to run. But the frontier model matches or beats it with zero training and no hosting. At a thousand calls the cost gap is pocket change. It only becomes a decision at scale.

The cost gap compounds at scale. Model it →
Fine-tuning worked here. It still isn't worth it.

Generic summarization is exactly the case where a frontier model wins. So when does fine-tuning earn its keep? There are three real answers.

The decision framework →
The fine-tuned model on Hugging Face ↗