Benchmarks & method
We publish the number, the conditions and the date. The 17 November 2026 run measured 7.5 ms p50 and 126 ms p95 time to first byte on cached anonymous HTML across 12 global regions. It is not a guarantee for every visitor.
What we measured
Three LumaWP sites on the production platform: a controlled fresh WordPress install and two live production sites. Every target was measured on cached anonymous pages with the same 12-region probe list.
| Target | p50 TTFB | p95 TTFB |
|---|---|---|
| Controlled test site (fresh WordPress) | 7.5 ms | 126 ms |
| Live production site 1 | 8.5 ms | 83 ms |
| Live production site 2 | 8 ms | 81 ms |
Same harness, reference hosts
We run the identical probe list against public WordPress sites on the hosts our customers compare us with. They are different sites with different content — measure them as orientation, not as a controlled experiment.
| Host | p50 TTFB | p95 TTFB |
|---|---|---|
| Cloudways | 8.5 ms | 29 ms |
| Hostinger | 10 ms | 15 ms |
| WordPress VIP | 28.5 ms | 49 ms |
| Rocket.net | 30.5 ms | 82 ms |
| Elementor Cloud | 46 ms | 145 ms |
| Pressable | 93.5 ms | 624 ms |
| SiteGround | 257 ms | 368 ms |
| WP Engine | 349–606 ms | 855–1,662 ms |
| Kinsta | 527 ms | 1,001 ms |
Reference measurements: 13 November 2026, the same 12 locations, each host’s own public site. WP Engine’s figures are the fastest of three public WP Engine sites measured in that run (349–606 ms p50).
PageSpeed Insights
On 18 November 2026 the canonical LumaWP host scored 93/100 desktop and 75/100 mobile. The mobile score is limited by the site’s own third-party analytics tags and stylesheet count — site content, not the hosting layer. Lighthouse simulations commonly move ±5–10 points between runs.
Method
| Measurement | Page time to first byte (TTFB) from the public hostname |
|---|---|
| Cohort | Cached anonymous HTML only; personalised, admin and API responses excluded |
| Probe locations | 12 global regions, the same location list for every host |
| Run | One probe per location, fresh TLS, no keep-alive reuse; p50 and p95 reported |
| Delivery | Global edge network; cached anonymous pages measured from the public hostname |
| Date | 17 November 2026 |
| Caveat | Internal benchmark, not a service-level agreement. Real results depend on location and cacheability. |
What this does and does not prove
It proves that cached anonymous pages are delivered quickly from the global edge under the measured conditions. It does not prove that every page, plugin or visitor will see the same number, and it is not a promise of “fastest” in every region or against every host. Personalised pages, admin requests and uncached API responses are always slower; we report them separately when we have a reliable method.
We re-run the benchmark when the platform changes. If a run regresses, this page shows the new numbers — the old ones are not quietly edited out.