Pages arrive like they do in the reference browser, in a fraction of the memory
This release rebuilds how the engine loads a page. We studied how the reference browser does it, from its source code and the web standards, measured eleven real sites side by side with the reference browser, and rebuilt our pipeline to the same model instead of tuning it on one site. The first picture of a page now waits only for what the reference browser waits for: the page itself, its stylesheets and the few small scripts at the top. Everything else (images, video, fonts, the rest of the scripts) streams in afterwards. A code-hosting site used to take about twenty seconds to show anything and now takes about one; a developer-docs site went from fifteen seconds to about two. Underneath, the network got a real upgrade. HTTP/3, the newest web protocol, now carries all of a site's requests over one connection at once, keeps working when a server rotates its encryption keys mid-connection, and asks for compressed files. The browser also got much lighter: a simple page used to need over two gigabytes of memory and now needs about 170 megabytes. Fonts are shared across tabs and the ad and tracker blocker is seven times smaller, with identical decisions. Scrolling got smoother too, most of all on Windows, where frames went from about 17 milliseconds each to about 3.