Skip to content
Today
Sections

Build Independently confirmed2 publishers 3 min read Published Updated

Lightpanda 1.0 makes a browser without a renderer a production option for crawlers and agents

Lightpanda shipped 1.0 of its open-source browser on October 2nd, passing 1,739,845 web-platform subtests without rendering pages. One encoding suite makes up about two-thirds of that count, so teams should judge it by whether their jobs need to see the page.

The Engineer · Build desk

Send a correction

What happened

  • Lightpanda passed 2,645 subtests when it entered beta in November 2024, and roughly 290,000 by March 2026.
  • Lightpanda puts its total at about 80% of Chrome's September 2026 count, with the gap concentrated in CSS, editing and SVG tests tied to layout and painting.
  • CORS enforcement is now on by default, a change aimed at a browser that runs code from sites its operator does not control.
  • DeveloperHub.io's founder says moving a prerendering service off headless Chrome made pages serve four times faster and cut average system load tenfold.
  • Vercel's agent-browser can switch to Lightpanda with a command-line flag, and Nous Research's Hermes Agent supports it as a local browser option.

Why it matters

  • Who has to decide decision Teams have to sort each job by whether any step needs to see the page, because screenshot, PDF and layout-dependent work still requires a full visual browser.
  • What it limits constraint The 1.7 million total cannot stand in for compatibility; take out the encoding suite and roughly 590,000 subtests remain to cover everything else in the browser.
  • Who's exposed exposure Jobs built on the beta whose page scripts fetch from other origins will hit Chrome-style restrictions after upgrading to 1.0 and need retesting.
  • What it costs cost Overage on the hosted service costs 8 cents an hour against about 6.3 cents for hours included in the $19 plan, so a 1,000-hour month bills $75.

Lightpanda runs a page's JavaScript and builds a DOM with the web APIs that driving code expects [2]. It leaves out graphical rendering, so it never produces pixels. The engine is open source and written in Zig [2]. It is built for crawlers, search and extraction systems, test suites and AI agents that need a page's content and a way to interact with it, and have no use for a screenshot [3].

The project started as a cost problem. Francis Bouvier, Lightpanda's co-founder and CEO, says it grew out of a crawler at his previous company that collected data from more than 20 million pages a day through headless Chrome [4]. The stated aim is to cut the compute and upkeep cost of running lots of Chrome instances when software, not a person, is doing the browsing [5].

The 1,739,845 figure needs unpacking before it goes into a planning document [1]. The encoding suite that started passing in April holds about 1.15 million subtests, many of them generated from character-mapping tables [7]. That one suite equals about 66% of the headline total [23]. For agent work, I'd look instead at the roughly 340,000 passes added from April through September in workers, Shadow DOM, IndexedDB, XPath, WebSockets and forms [8]. According to Runtimewire, the raw total counts each test the same no matter which part of the browser it checks [9].

The comparison with Chrome needs the same care. Taken at face value, the 80% figure puts Chrome's September count near 2.17 million and leaves a gap of roughly 435,000 subtests [25]. That gap sits in the layout and painting work an engine without a renderer skips [11]. Lightpanda does not render pages, take real screenshots or target PDF rendering [12]. So the question to ask of each job is whether any step needs to know what the page looks like. A step that does has to stay on a full visual browser [12].

DeveloperHub's result is a claim about one prerendering service [13]. For it to carry over, a job needs the property Runtimewire identifies as what Lightpanda offers: the output is page content, and nobody needs rendered pixels [14]. Keenable's figure covers the other half of the case. It says more than 35% of its AI-search answers depend on JavaScript-rendered pages, and that Lightpanda makes that work scalable [15]. Lightpanda says some users process more than 30 million pages a day, but did not name them or give a breakdown [16]. These are company- and partner-reported figures, and none has been independently audited [17].

I think turning CORS on by default is the right setting for an engine whose daily work is running JavaScript written by strangers [18]. The change does not affect page navigation [19]. The team recommends keeping the control enabled [20].

What to watch

  • An independent benchmark of Lightpanda against headless Chrome on a published workload, which would show whether DeveloperHub's fourfold speedup holds outside prerendering.
  • Pass counts in the CSS, editing and SVG suites, where the roughly 435,000-subtest gap to Chrome sits.
  • Whether Lightpanda names the users it says process more than 30 million pages a day, or breaks that figure down.

Clarity's read

What the record supports and how the coverage leans. The claims behind it follow.

Reality

Evidence55
Adoption35
Hype gap+25
Incentives70
Confidence60
Why these scores

Claim ledger

Ranked by verification strength, evidence, and original report placement.

  1. [1]

    Lightpanda released version 1.0 on October 2nd, ending the beta period; the company says the release passes 1,739,845 Web Platform Tests subtests.

    Reported Supported Source: Lightpanda, via RuntimewireView cited source
  2. [2]

    The open-source browser, written in Zig, runs JavaScript and provides a DOM and web APIs, but omits graphical rendering.

    Reported SupportedView cited source
  3. [3]

    The design targets crawlers, search and data extraction systems, tests, and AI agents that need page content and interaction rather than screenshots.

    Reported SupportedView cited source
  4. [4]

    Francis Bouvier, Lightpanda co-founder and CEO, says the project grew from operating a large-scale web-crawling system at his previous company, which collected data from more than 20 million pages a day using headless Chrome.

    Reported Supported Source: Francis BouvierView cited source
  5. [5]

    That experience informed Lightpanda's effort to remove the compute and maintenance overhead of running many Chrome instances for machine-driven browsing.

    Reported SupportedView cited source
  6. [6]

    Lightpanda's own account puts the browser at 2,645 subtests when it entered beta in November 2024, then roughly 290,000 by March 2026.

    Reported Supported Source: LightpandaView cited source
  7. [7]

    In April the encoding test suite, which contains about 1.15 million subtests, many built from generated character-mapping tables, began passing.

    Reported SupportedView cited source
  8. [8]

    Lightpanda says it added roughly 340,000 further passes from April through September, working on areas including workers, Shadow DOM, IndexedDB, XPath, WebSockets and forms.

    Reported Supported Source: LightpandaView cited source
  9. [9]

    The raw total gives equal weight to tests that exercise very different parts of a browser.

    Reported Supported Source: Runtimewire analysisView cited source
  10. [10]

    Lightpanda puts its total at about 80% of Chrome's September 2026 WPT count.

    Reported Supported Source: LightpandaView cited source
  11. [11]

    The remaining gap to Chrome is concentrated in CSS, editing and SVG tests tied to layout and painting, according to the release post.

    Reported Supported Source: Lightpanda release postView cited source
  12. [12]

    Lightpanda does not render pages, take real screenshots or target PDF rendering; teams that need those capabilities still need a full visual browser.

    Reported SupportedView cited source
  13. [13]

    DeveloperHub.io's founder says moving a prerendering service from headless Chrome to Lightpanda made pages serve four times faster and cut average system load tenfold.

    Reported Supported Source: DeveloperHub.io founder, customer-reportedView cited source
  14. [14]

    The operational trade Lightpanda is selling is less infrastructure for workloads that do not need rendered pixels.

    Reported Supported Source: Runtimewire analysisView cited source
  15. [15]

    Keenable says more than 35% of its AI-search answers depend on JavaScript-rendered pages and reports that Lightpanda makes the work scalable.

    Reported Supported Source: KeenableView cited source
  16. [16]

    Lightpanda's release announcement says some users process more than 30 million pages a day, without identifying them or providing a breakdown.

    Reported Supported Source: LightpandaView cited source
  17. [17]

    The usage examples are company- and partner-reported, not independently audited deployment figures.

    Reported Supported Source: RuntimewireView cited source
  18. [18]

    The release turns CORS enforcement on by default, a change that matters when a browser executes code from sites an operator does not control.

    Reported SupportedView cited source
  19. [19]

    Lightpanda says browser navigation remains unaffected, while cross-origin fetch requests initiated by page scripts now follow the same restrictions as Chrome.

    Reported Supported Source: LightpandaView cited source
  20. [20]

    The Lightpanda team recommends keeping the CORS control enabled.

    Reported Supported Source: LightpandaView cited source
  21. [21]

    Vercel's agent-browser can select the Lightpanda engine with a command-line flag, and Nous Research's Hermes Agent supports it as a local browser option.

    Reported SupportedView cited source
  22. [22]

    Lightpanda's hosted cloud pricing page lists a free plan with 10 browser hours a month and a $19 monthly plan with 300 hours, then $0.08 per additional hour; high-volume and private-cloud terms are custom.

    Reported SupportedView cited source
  23. [23]

    The encoding suite's roughly 1.15 million subtests equal about 66% of the 1,739,845 total.

    Derived Supported
  24. [24]

    Subtracting the encoding suite's size from the total leaves roughly 590,000 subtests for the rest of the browser.

    Derived Supported
  25. [25]

    If Lightpanda's total is 80% of Chrome's, Chrome's September 2026 count is roughly 2.17 million and the gap is roughly 435,000 subtests.

    Derived Supported
  26. [26]

    On the $19 hosted plan, included hours cost about 6.3 cents each against 8 cents for overage, and a 1,000-hour month costs $75.

    Derived Supported

Sources

2 independent publishers whose own reporting we read for this story.

  1. lightpanda.io

    1 article · October 6, 2026

    Lightpanda 1.0: A Headless Browser for Automation
  2. runtimewire.com

    1 article · October 2, 2026

    Lightpanda ships 1.0 after passing 1.7 million web-platform tests

Topics and entities

Topics

Entities