If your PageSpeed Insights score changed in the last week and nobody touched the page, the likely reason is Lighthouse 13.6. PageSpeed Insights now runs that version, which changes how it estimates load times on a simulated phone. Google's Barry Pollard said on Nov. 2 that it gives more accurate, "usually better" scores.
We wanted to know how much better, and whether it is always better. So we took single page loads from 11 public sites and scored each one with both Lighthouse 13.5.0 and 13.6.0. Same page load, same data, only the version changed. Scores moved by between -4 and +9 points. Ten of 22 went up, four went down and eight didn't move.
What changed in Lighthouse 13.6
Lighthouse 13.6.0 was released on GitHub on Oct. 27, 2026. Its lab scores don't come from a real slow phone. Lighthouse loads the page on a fast connection, then simulates how long it would have taken on a slower one. Version 13.6 changes that simulation. The release notes list, among others:
- more accurate request and network modeling, including treating HTTP/3 connections as multiplexed
- using Chrome's own render-blocking label to decide which requests hold up the first paint
- using the first paint's timing for Largest Contentful Paint "when text LCP occurs in the same frame as FCP"
- a fix to the simulated LCP breakdown
Pollard said 13.6 is "now on PSI" and will reach Chrome DevTools in Chrome 159, and that it is "more likely to impact JS-heavy sites."
How we tested it
Lighthouse can save a page load (the trace and network log) and score it later. That let us remove the biggest source of noise in speed tests, which is the page load itself.
- Pages: the home pages of The Search Post, react.dev, CNN, The Verge, Target, Shopify, BBC News, Zillow and WordPress.org News, plus Google's SEO Starter Guide and Wikipedia's article on search engine optimization. Etsy blocked our test browser, so it isn't included.
- Loads: two separate loads of each page on Nov. 9, from a cloud server, using Lighthouse's default mobile settings (the same simulated phone PageSpeed Insights uses for its mobile score) in headless Chrome 153.
- Scoring: each saved load scored once by Lighthouse 13.5.0 and once by 13.6.0, performance category only.
That gives 22 pairs. Our absolute scores won't match what PageSpeed Insights shows for these pages, because it runs from Google's servers. The difference between versions is the part that carries over.
What we found
| Page | Score, load 1 (13.5 to 13.6) | Score, load 2 (13.5 to 13.6) | LCP change, loads 1 and 2 |
|---|---|---|---|
| react.dev | 82 to 91 | 78 to 87 | -2.6 s, -2.8 s |
| Target | 19 to 25 | 20 to 26 | -0.7 s, -0.4 s |
| Google SEO Starter Guide | 42 to 47 | 58 to 64 | -0.6 s, -0.9 s |
| WordPress.org News | 93 to 93 | 94 to 96 | -0.1 s, -0.7 s |
| BBC News | 57 to 57 | 45 to 47 | none, -0.3 s |
| The Verge | 34 to 35 | 32 to 33 | -0.2 s, -0.5 s |
| Wikipedia article | 69 to 69 | 99 to 99 | -0.3 s, -0.4 s |
| The Search Post | 91 to 91 | 91 to 91 | none, none |
| Zillow | 36 to 35 | 35 to 35 | none, -0.2 s |
| Shopify | 66 to 65 | 64 to 64 | none, none |
| CNN | 34 to 30 | 33 to 31 | -0.3 s, -0.6 s |
Three patterns stood out:
- Largest Contentful Paint never got worse. It dropped in 16 of 22 pairs and stayed the same in the other six. The biggest drop was react.dev, by 2.6 and 2.8 seconds.
- On some pages LCP now equals the first paint. In seven pairs (react.dev, Wikipedia, Google's guide, and one WordPress.org load), 13.6 put LCP at exactly the same time as First Contentful Paint, where 13.5 had it later. That matches the release note about text LCP painted in the same frame.
- Total Blocking Time went up on heavy pages. On CNN it rose by about 2 seconds in one load and 1.6 seconds in the other, and on Target by about 1.5 seconds in both. On CNN that outweighed faster paint times, so its score went down. Lighthouse 13.6 also moved First Contentful Paint earlier on these pages (by 2.1 to 2.9 seconds on CNN). Total Blocking Time only counts script work after the first paint, so an earlier first paint likely puts more of it inside the window.
Our own site, a static page with little JavaScript, came out identical on both versions to the millisecond. That fits Pollard's point that JavaScript-heavy sites are the ones likely to move.
The bigger lesson: one run means little
The difference between two loads of the same page, on the same version, was often bigger than the difference between versions. Wikipedia scored 69 on one load and 99 on the other, because a layout shift happened in the first load and not the second. BBC News went from 57 to 45 between loads. If a single PageSpeed Insights run moved your score by a few points, that may not be the version or your site at all.
Why did my PageSpeed score go up?
If it went up in late October or early November and you didn't ship a change, Lighthouse 13.6 is the most likely reason, especially on a JavaScript-heavy site. Check the Lighthouse version at the bottom of the report. Before you compare an old report with a new one, make sure both say 13.6.0. A higher score from a version change is not a faster site.
PageSpeed Insights lab data vs field data
None of this touches the field data at the top of the report: the Core Web Vitals from real Chrome users (the Chrome UX Report). That is what Google uses for its page experience signals, not the lab score. If your field data didn't move, your visitors' experience didn't either.
Run the same check on your own page
You need Node.js and Chrome. This loads your page once, then scores that same load with both versions:
mkdir lh-old lh-new
(cd lh-old && npm init -y >/dev/null && npm i [email protected])
(cd lh-new && npm init -y >/dev/null && npm i [email protected])
URL=https://example.com/
# Load the page once and save the trace (-G = gather only)
lh-old/node_modules/.bin/lighthouse $URL --only-categories=performance -G=./load
# Score that same load with each version (-A = audit saved data)
lh-old/node_modules/.bin/lighthouse $URL --only-categories=performance -A=./load --output=json --output-path=old.json
lh-new/node_modules/.bin/lighthouse $URL --only-categories=performance -A=./load --output=json --output-path=new.jsonThen compare categories.performance.score and the largest-contentful-paint and total-blocking-time audits in the two files. Do it for two or three loads before you trust the result.
What we don't know yet
- How PageSpeed Insights' own numbers moved. Google hasn't published a before and after across sites, and our test ran from our server, not Google's.
- Whether 13.6 is more accurate for your pages. Pollard says the simulation is more accurate. The real check is your field data, which only comes from actual visitors.
- When Chrome DevTools users see it. Pollard says it will ship in Chrome 159, which was still in Canary on Nov. 2.
