Disclosure: we make Y Build. So we published our full method, raw timings, and the exact prompts — reproduce it yourself. The numbers are what we’d want to see before trusting a vendor’s own benchmark.
The experiment
“Vibe coding” tools all promise the same thing: describe an app, get an app. But the demos always stop at a working preview — never at a deployed, real product with a database and a URL you can send to a customer. So we ran the boring version of the test.
We took one spec and built it three times — in Bolt.new, Lovable, and Y Build — and timed everything with a stopwatch and a screen recorder.
What we built
The spec: a bookings app — sign-in, a calendar, a Postgres-backed appointments table, an admin list, and a public booking page on a custom domain. Identical prompt for all three, written out below.
Build a bookings app. Users sign in with email. A calendar shows open 30-min slots. Booking writes to a Postgres `appointments` table. Add an admin page listing today's bookings. Deploy it to a public URL on a custom domain.
Results: prompt → live URL
All three reached a working preview fast. The gap opened at deployment — the part the demos skip. Here’s the full scorecard.
| Metric | Bolt.new | Lovable | Y Build |
|---|---|---|---|
| Time to working preview | 18 min | 22 min | 14 min |
| Prompts to reach MVP | 9 | 12 | 7 |
| Extra time to live URL | +12 min | +9 min | +3 min |
| Manual config steps | 6 | 4 | 0 |
| Cost for this build | $20/mo | $25/mo | $12/mo |
Cost over 30 days
Preview pricing hides the real bill: databases, hosting add-ons, and the seat you need to keep it live. Over a month of light traffic, the spread widened.
“The build was a tie. The deploy was the whole story — and it’s the part nobody benchmarks.”
Where each tool wins
None of these is bad. They’re tuned for different jobs:
- Bolt.new — best for full-stack prototyping when you’ll deploy elsewhere.
- Lovable — prettiest first-pass UI; great for design-led MVPs.
- Y Build — fewest steps to a real, deployed product on your domain.
Methodology
Same machine, same network, fresh accounts. One operator, three runs per tool, stopwatch started at first prompt and stopped when the public URL resolved over HTTPS. We took the median to dampen luck. Raw screen recordings and prompt logs: github.com/ybuild/lab-03.
The verdict
If you’re prototyping, pick the one whose UI you like. If you’re shipping a real product to a real domain, the deploy gap is decisive — and it’s why we built Y Build the way we did.