JavaScript framework popularity in 2026: four sources, four different answers

HTTP Archive counts origins on which a framework was detected. July 2026 data; the full population for the technology report is 8,877,469 origins:

The share column is our arithmetic — HTTP Archive publishes counts only and states no percentage. The percentages shown beside each technology on their dashboard are month-over-month change, not market share; the two must not be confused.

Year-on-year from the same data (July 2025 → July 2026): React 1,180,640 → 1,306,221; Vue 520,219 → 609,907; Next.js 231,551 → 357,931; Svelte 74,130 → 90,479.

One figure in that table is unreliable. Angular jumped from 116,639 to 217,284 origins between March and April 2026. The source offers no explanation; organic growth does not look like that, and presenting it as a doubling in popularity would be wrong.

W3Techs as of 27 August 2026, two columns — of all sites, and of sites where any library was detected:

On jQuery the sources essentially agree: 66.5% at W3Techs against 65.0% in our arithmetic on HTTP Archive. That is a useful checkpoint — two independent detectors on different samples landing in the same place.

But on Angular the gap is tenfold: 214,013 origins at HTTP Archive against 0.2% at W3Techs, while BuiltWith gives a third answer — 52,737 live sites on a population four times larger. No source explains it. The right conclusion is not to average them but to accept that Angular detection is unreliable: it leaves no fingerprint as unambiguous as React’s or jQuery’s.

This is the easiest place to go wrong, because the numbers look comparable while the quantity is different — not "runs on sites" but "a person answered so".

Three columns, three different questions. State of JS (fielded September–November 2025, 13,002 responses) asks whether a respondent has ever used an item — cumulative over a career; the average respondent named 2.6 frameworks. Stack Overflow (fielded May–June 2025, 49,009 responses, 23,678 on web frameworks) asks about extensive work over the past year. W3Techs looks at sites and knows nothing about people.

Both surveys are self-selected, recruited through community channels. State of JS says outright that its results "should be taken as a snapshot of a specific subset of developers, and is not meant to speak for the entire ecosystem".

A separate dating trap. Stack Overflow Developer Survey results for 2026 do not exist — the survey only opened on 23 June 2026. Nor is there a State of JS 2026. Any article citing "Stack Overflow 2026" framework percentages is recycling 2025 or inventing.

Over the 30 days from 27 July to 25 August 2026: react 669,873,506; express 540,562,648; next 224,320,750; preact 123,659,539; jquery 72,831,123; vue 59,104,805; @angular/core 24,676,258; svelte 23,032,239.

This is a count of requests to the registry and nothing more. CI builds, Docker images and mirrors all sit inside it: a single project can generate thousands of downloads a day. Transitive dependencies inflate the figures — which is exactly why preact outranks jQuery here despite running on fifteen times fewer sites.

The skew runs the other way too: jQuery and htmx are more often included via a CDN tag, bypassing npm entirely, while Angular publishes across a dozen packages, so @angular/core understates it.

The takeaway: npm downloads convert into neither sites nor developers. They are a separate quantity, useful only for comparing a package against itself over time.

To check what a specific site loads, use the technology detector.