Organisation pour le Développement de l’Éducation, des Ressources et des Savoirs AccessiblesNonprofit association · Paris

Technical proof

Speed, measured from 6 continents

We do not ask anyone to take our performance on trust. Here is the whole battery: 78 measurements taken by physical probes in 14 cities across 12 countries, the raw file to download, the method to run it again, and what these figures do not prove.

Measured on 31 July 2026 · 78 measurements · 6 sites in the fleet · 14 cities · 12 countries · 6 continents

The fleet, in four figures

The fleet, in four figures

94 msmedian response time, across every single measurement
76/78measurements under Google’s good threshold, which is all of them
57/78measurements under 200 ms, the stricter bar we set for ourselves
0requests to a third-party server, on every page we publish

Battery of 31 July 2026, taken with the public Globalping probes. The raw file sits at the foot of this page.

On the real web, 44% of sites meet Google’s good threshold for response time on mobile. In our battery: 76 measurements out of 78.

Web distribution: HTTP Archive Web Almanac 2025, from Chrome UX Report data of July 2025. Our figures: battery of 31 July 2026, raw file at the foot of this page.

47 msmedian response time when the page is already cached in the visitor’s own region, which covers 61 of our 78 measurementsRun the measurement again →
17 msfastest measurement in the battery, across every site and every city
8145 msslowest measurement in the battery, empty cache and slow probe included: we publish it as measured rather than set it aside
42/78measurements under one hundred milliseconds, the point beyond which a response stops feeling instant
100desktop performance score on PageSpeed Insights, alongside accessibility, best practices and searchRun the test again →
0data collected, so there is no audience measurement script to load before the page can appearSee how to check it →

Why it is fast

Why it is fast

Pages already written

Our sites are files, not programs. There is no database to query and no template to assemble when a visitor arrives: the page requested already exists, in full, before anyone asks for it. The server only has to hand it over.

Served from the city next door

Every page is copied across our host’s worldwide network and served from the point of presence nearest the visitor. A student in Mumbai does not download Paris: she downloads Mumbai. That is exactly what the 14 cities on this page measure.

Nothing loaded from anyone else

No font, no script, no image and no audience measurement comes from another domain. A page that depends on nobody cannot be slowed down by anybody, and checking it takes thirty seconds in any browser’s network inspector.

Light by construction

Text first, photographs given their dimensions, fonts carried with the page, and the page stays readable even if the script never runs. Speed here is no feat of engineering: it is what remains once nothing useless has been added.

What this figure measures, and what it does not

Every measurement on this page is a response time: the delay between the browser asking for the page and the first byte of the answer arriving. It is the foundation, not the house.

What it includes
Resolving the domain name, opening the connection, negotiating encryption, then waiting for the answer. Across our battery the median connection opens in 2 ms and the median encryption takes 11 ms.
What it leaves out
Downloading the rest of the page, drawing it, laying it out. A short response time does not guarantee a page that feels fast, which is why we publish Google’s scores as well: those measure what appears on screen.
Why Google’s threshold sits at 800 ms
Because beyond it the first paint no longer fits inside the budget Google treats as good. The threshold is read at the 75th percentile of real visits: it is not enough for half the measurements to pass, three quarters must.
Why we aim for 200 ms
Because that is the bar commonly quoted for excellent infrastructure, and it suits us better: 57 of our 78 measurements clear it. It is no longer an official Google threshold, and we would rather write that down than lay claim to it.

Response time, continent by continent

Each row gathers the measurements taken on that continent, across all of our sites, fastest first. The median is the value that splits the measurements into two equal halves.

Response time, continent by continent
ContinentCities probedMeasurementsMedianFastestSlowest
Europe31839 ms17 ms486 ms
North America31849 ms21 ms102 ms
Oceania16116 ms21 ms285 ms
South America16139 ms28 ms151 ms
Asia424233 ms21 ms8,145 ms
Africa26252 ms30 ms268 ms

The continent is derived from the country of the probe. Not one figure in this table is written by hand: they are all recalculated from the raw file every time the site is published.

Site by site, across the fleet

The 6 sites the association publishes, measured in the same pass. The median shown here is the one for answers already cached: the ordinary visitor’s experience.

Site by site, across the fleet
AddressMeasurementsCachedCached medianFastestSlowest
odersa.org1311/1324 ms17 ms497 ms
letabli.odersa.org136/1340 ms27 ms449 ms
depleindroit.odersa.org1311/1348 ms25 ms790 ms
avantlessecours.odersa.org1311/1350 ms22 ms2,723 ms
clicparclic.odersa.org1312/1356 ms21 ms8,145 ms
avantdycroire.odersa.org1310/13116 ms25 ms791 ms

The spread between our sites comes from how often they are visited, not from their code: a page rarely requested in a region is not cached there yet. Between 24 and 116 ms of median, all of them stay far below Google’s threshold.

The full record, city by city

The 14 cities of the battery, each with its measurements across our 6 sites. This is the least flattering of the three tables, which is exactly why it is here: nothing has been removed.

The full record, city by city
CityContinentMeasurementsCachedMedianFastestSlowest
LondonEurope66/625 ms23 ms64 ms
New YorkNorth America66/632 ms24 ms50 ms
FalkensteinEurope66/640 ms29 ms486 ms
WarsawEurope66/646 ms17 ms117 ms
TorontoNorth America66/657 ms34 ms97 ms
TokyoAsia65/664 ms21 ms267 ms
Los AngelesNorth America66/681 ms21 ms102 ms
SingaporeAsia64/6113 ms28 ms215 ms
SydneyOceania65/6116 ms21 ms285 ms
Sao PauloSouth America65/6139 ms28 ms151 ms
JohannesburgAfrica54/5248 ms30 ms258 ms
Cape TownAfrica11/1268 ms268 ms268 ms
MumbaiAsia60/6269 ms215 ms335 ms
DubaiAsia61/6791 ms449 ms8,145 ms

A city where few measurements were cached shows a higher median: that is the first visitor in the region waiting, once, while the page settles in nearby.

Our measurements against public benchmarks

The gap column compares our median of 94 ms with the benchmark on that row. None of these benchmarks are ours: each carries its source, and the address opens in one click.

Our measurements against public benchmarks
Public benchmarkBenchmark valueGap with our medianSource
Google’s good threshold for response time800 ms× 9web.dev (Google)
Google’s poor threshold, beyond which a site is flagged1,800 ms× 19web.dev (Google)
Bar commonly quoted for excellent infrastructure200 ms× 2Auditite, 2026 benchmarks
Median of the fastest sector in that record: finance400 ms× 4Auditite, 2026 benchmarks
Median of the healthcare sector600 ms× 6Auditite, 2026 benchmarks
Median of the media sector750 ms× 8Auditite, 2026 benchmarks

Stated reservation: the source of these sector medians publishes neither its method nor its sample, and it lists sectors beyond these three. We quote it as an order of magnitude, never as proof. The 200 ms bar is no longer an official Google threshold either: it comes from documentation withdrawn in 2019. Our own figures, by contrast, recalculate in front of you from the raw file.

Google’s public scores

PageSpeed Insights is Google’s public tool: anyone can run it on our pages and on other people’s, then compare. Recorded on desktop on 31 July 2026.

Google’s public scores
What is measuredDesktop result
Performance100 / 100
Accessibility100 / 100
Best practices100 / 100
Search100 / 100
Largest element painted0.4 s
Layout stability0.001

We are not copying a mobile score here. A score has a date, and ours has just been corrected for a stability fault specific to phone rendering: the honest way to know it is to run the tool, not to read us.

Run PageSpeed Insights on our homepage

Run the measurements yourself

Everything on this page can be reproduced without us. Both tools are public and free, and neither asks for an account.

  1. Open the Globalping probe service, pick the HTTP test, enter odersa.org as the address and the whole world as the origin: physical probes spread across the planet will query the site and show their own timings, just like ours.
  2. Open PageSpeed Insights, paste the address of one of our pages, and read the four scores and the paint measurements. Do the same on another site to give yourself a point of comparison.
  3. Open any page of ours, press F12, go to the Network tab and reload: every request carries our own domain, and none leaves for anywhere else.
  4. Download the raw file at the foot of this page and recalculate our medians: you should find the figures in our tables, row for row.

A repeated measurement will not return exactly ours: the probes available change from hour to hour and each one’s network varies. That is what an honest measurement looks like. The orders of magnitude, however, hold.

Our reservations, written before anyone raises them

A measurement without its limits is an advertisement. Here are the limits of ours.

  • These figures are a snapshot from 31 July 2026, not a permanent guarantee. Unlike the figures about our content, which recount themselves at every publication, a network measurement means running the probes again.
  • 17 of our 78 measurements found the regional cache empty: that is the first visitor in a region waiting, with a median of 270 ms instead of 47 ms. Once. We checked it live on Sydney, where the second request came back in line with the rest.
  • Share of the battery coming from probes whose own connection was slow: 1 out of 78. It is readable in the raw file, where their connection alone takes more than one hundred milliseconds, before we have answered anything. We keep them in the tables rather than setting them aside.
  • A probe is not a visitor. It measures a network and a server, not a second-hand phone on a congested network. Google’s scores cover that part, and Google measures them, not us.
  • Speed is not teaching. A fast site that teaches badly is still a bad site. This page proves one thing and one thing only: that our infrastructure is not in the way.
  • We publish no visitor counts, because we collect nothing that would let us count them. What you read here measures our servers, never our readers.

The raw record, as open data

The file our tables summarise, exactly as it came out of the probes: one row per measurement, with the site queried, the city, the cache state and the detailed timings. Under the CC BY 4.0 licence, like everything else we publish.

Download the record (78 measurements)

Real work, free, and verifiable in a click

Partnerships on the ground, support, press: one door, and a reply from a real person.

Write to us

Keyboard shortcuts

TabMove from link to link
EnterOpen the link, or expand and collapse a question
?Open this help
EscClose the panel or this help