Break-even board
Set your own task profile — monthly query volume, required accuracy, latency budget — and three cost curves for prompting, RAG, and fine-tuning plot against volume with a marker on exactly where your numbers land. Drag the accuracy bar and the "right" strategy visibly flips. It replaces fine-tune-vs-prompt hot takes with your own crossover math. Runs entirely in your browser (0 uploads, works offline).
How to use it
- Set your monthly volume, accuracy requirement, and latency budget.
- Watch the three cost lines and where your volume marker crosses them; disqualified strategies (too low accuracy, too slow) are flagged.
- Read the recommendation and the X-ray: the cheapest qualifying strategy for your numbers, and the exact volumes where the answer flips.
What this clears up (the fundamentals)
- It's a crossover, not a rule — each strategy trades fixed cost (setup / training) against per-query cost. Prompting has no fixed cost but pays the most per call; fine-tuning is the reverse.
- Volume decides the winner — at low volume, prompting's zero setup wins; as volume rises, RAG then fine-tuning earn back their fixed cost through cheaper queries.
- Requirements gate the options — accuracy and latency ceilings can disqualify a strategy entirely, flipping the answer regardless of cost.
- The answer is personal — the same "just fine-tune it" advice is right for one team and wrong for another; only your own volume and requirements settle it.
Where it's used
An AI build-vs-buy decision tool — the fastest way to cut through "you should fine-tune" / "just prompt it" debates with a number that's actually yours. It's a miatz build-lab concept playable; learn to build it yourself.
FAQ
How is each strategy's cost computed?
Fixed cost plus per-query cost times volume. Prompting has zero fixed cost but the highest per-query price; RAG adds a setup cost for a cheaper query; fine-tuning has the largest fixed (training) cost and the cheapest queries.
What does "disqualified" mean?
A strategy is removed from the running if it can't reach your required accuracy or can't respond within your latency budget — even if it would be the cheapest, it doesn't qualify.
What's a crossover volume?
The volume at which two strategies cost exactly the same. Below it, the lower-fixed-cost option is cheaper; above it, the lower-per-query option pulls ahead. The tool marks the two key crossovers.
Is anything uploaded?
No. All the cost math runs in your browser — nothing is transmitted or stored. Turn off your Wi-Fi and it still works.
Limits
A teaching model with three preset strategies and linear costs — real decisions include quality variance, maintenance, vendor lock-in, and data-availability constraints. Use it to build the crossover instinct, then plug in your own vendor quotes.
Related
Part of the Demystify Playgrounds. Explore the rest from the Playgrounds home.
Bookmark this page (Ctrl+D, or ⌘D on Mac) — it works offline the next time you need it.