Skip to content
Deploy Bench

Day zero

The benchmarks are not
written yet. The rules are.

This is a deploy benchmark being built in public. The measurement rules go up first, the hosts go into a queue, and a figure only reaches this site once there is a run behind it that I can show you.

Bench log

Nothing has been measured yet.

Six hosts are queued and none of them has been run. Publishing the queue at zero is the point: a comparison written before the runs exist is a shopping list. The table fills in as each run completes, and until then it stays empty.

Hosts queued for benchmarking. None has been run yet.
HostTypeStatus
Cloudflare PagesStatic / edgemeasured
Kinsta StaticStaticqueued
CloudwaysManaged VPSqueued
WP EngineManaged WordPressqueued
DigitalOcean App PlatformPaaSqueued
HostingerShared / VPSqueued

How it will be measured

Four rules, fixed before the first run.

Every benchmark published here will have to satisfy these. A run that breaks one of them gets thrown away instead of published with a caveat attached.

01

One project, deployed everywhere

One reference repository goes to every host in the comparison, at the same commit, with the same lockfile and the same content. No host will get a hand-tuned build.

02

Cold and warm stay separate

First byte after a cache purge and first byte on a warm edge are different products. Both will be published as two numbers, never averaged into one.

03

No run, no number

A figure may only be published together with the command that produced it, the date it was run, and the raw output. Anything I cannot show the log for stays off the page.

04

Prices are read off the checkout page

Plan pricing will be read from the provider's own pricing page and printed with the date it was read. Introductory rates and renewal rates count as two separate numbers.

Who writes this

One working developer.

Deploy Bench is written by Taiga Kudo, who builds and ships client sites for a living. There is no editorial team here and no sponsored placement.

Which hosts get benchmarked is decided by the queue above, and how they get benchmarked is decided by the rules. Both were published before any run, so you can hold the results against them later.

  • Claims will ship a command

    Paste it into your own terminal and you should land on the same number.

  • Prices will carry a date

    Hosting pricing moves. Every figure will say when it was read off the provider’s own page.

  • Cold and warm stay two numbers

    Averaging them together is how a slow origin gets sold as a fast one.

Migration service

Or skip the reading and hand it over.

You send the repository and the host you are on now. You get back a working deployment on the new host, with the build pipeline, environment variables, DNS and TLS all configured. If it does not work to the agreed scope, you get your money back.

See what is included

$499 flat, per site