Lighthouse 13.6 changes how PageSpeed Insights simulates a page load

Lighthouse 13.6.0 came out on October 27, and its release notes said it would reach PageSpeed Insights "within a week" and Chrome DevTools with Chrome 159. The change that matters for anyone tracking scores is in Lantern, the model Lighthouse uses to simulate a page loading on a throttled phone. The team upgraded its trace engine and the request and dependency-graph modeling "to improve simulation accuracy".

What it does to your scores

Lab metrics such as Largest Contentful Paint and Total Blocking Time come out of that simulation. Change the model and the same page, unchanged, can score differently.

Before you hunt for the deploy behind a jump or a dip in early November, check which Lighthouse version produced each report.

Every report shows the Lighthouse version it ran on, at the bottom of a PageSpeed Insights report and in the JSON as lighthouseVersion. Treat 13.6 as a new baseline and don't chart old and new runs as one series.

Field data doesn't move. The real-user section of the report comes from the Chrome UX Report, which is also what Core Web Vitals in Search Console report on. A lab change touches neither.

The other changes in 13.6

• The CLI can audit several URLs one after another in a single command.
• Performance insight audits now report a numeric value and unit, taken from their estimated time or byte savings.
• Peak memory is lower, because traces are streamed during a run.
• The Agentic Resource Discovery audit follows version 0.91 of its spec.

If you track speed across many pages, the bulk CLI mode does in one command what used to need a script looping over URLs. For crawlers that run Lighthouse on a whole site, see the site audit tools in the directory.