Lightpanda 1.0 Passes 80% of Chrome's Web Tests, by Its Own Count
Software / news
Lightpanda 1.0 Passes 80% of Chrome's Web Tests, by Its Own Count
The Zig-built headless browser left beta on Oct. 2, and its own benchmarks put it ahead of two Rust rivals on a 940-page crawl, with no hardware listed.
Lightpanda, a headless browser written in Zig for AI agents and scrapers, declared 1.0 on Oct. 2, 2026. Cofounder and chief executive Francis Bouvier wrote that it "is no longer in beta" and that "you can depend on Lightpanda in production."
Every benchmark below was run by Lightpanda. The one independent set of numbers in this story is Cloudflare's, for a different browser.
What 1.0 claims
Bouvier's 1.0 post sets two conditions: enough web platform coverage for real sites to work, and security rules such as CORS enforced by default. On the Web Platform Tests, the shared standards suite browsers are measured on, Lightpanda passes 1,739,845 subtests, up from 2,645 in November 2024.
The post lists Chrome at 2,184,491 and Firefox at 2,137,997. Lightpanda's count is about 80 percent of Chrome's. In the cors/ suite it passes 371 of 463 subtests, and the remaining 92 are still in progress.
| Browser | WPT subtests passed | Source |
|---|---|---|
| Chrome | 2,184,491 | Lightpanda 1.0 post |
| Firefox | 2,137,997 | Lightpanda 1.0 post |
| Lightpanda | 1,739,845 | Lightpanda 1.0 post |
Most of the gap sits in the CSS, editing and SVG suites, which test layout and painting. Lightpanda has no graphical rendering. It does not take screenshots: --dump png and --dump pdf render the page's markdown, not the page as seen.
That limit decides who can use it. A crawler that extracts text from 940 pages loses nothing. A test suite that asserts on how a button looks cannot run here.
The 940-page crawl
On Oct. 19, Lightpanda cofounder and chief technology officer Pierre Tachoire published a comparison with three other browsers built from scratch. A script followed every link of a 940-page demo site, starting a fresh browser each run with 25 tabs.
- Lightpanda109 MB
- Obscura201 MB
- Moli230 MB
Source: Lightpanda blog, Oct. 19, 2026, accessed 2026-11-01
Lightpanda finished in 4.5 seconds, Obscura in 11.5 and Moli in 17.4. CPU time was 12, 29 and 40 seconds.
The post does not state the hardware, operating system or machine specifications for any test. The post is written by Lightpanda's own chief technology officer, and Lightpanda leads most of its comparisons.
The README's headline claim is a different test: "16x lighter and 9x faster than Chromium," from a table of 100 pages with 123MB of peak memory against 2GB and 5 seconds against 46. The README says the run requested 933 pages on an AWS EC2 m5.large. The table and the page count do not match, so the two figures are not directly comparable.
What the rivals and the licence look like
The same post tabulates the licences. Lightpanda is AGPL-3.0, a network-copyleft licence. Obscura, written in Rust, is Apache-2.0. Moli, also Rust, is Apache-2.0 or MIT. Kitesurf is closed source.
The post also says Obscura passes 652,000 WPT subtests (30 percent of the full suite) in the authors' run, while Obscura's own README reports 67.6 percent on a subset. Moli had not finished a complete run after 10 hours. Kitesurf's score is computed from published data because Lightpanda could not reproduce it.
The README lists CDP (the Chrome DevTools Protocol used by Puppeteer and Playwright), WebDriver BiDi, an MCP server and a fetch --dump command. It says builds need Zig 0.17.0, there is no native Windows or Android build, and musl-based Linux is unsupported. The repository shows 39.3k stars.
Cloudflare's number for a rival
Cloudflare's Kitesurf is the useful contrast, because the company published its own test and a developer summarised it. In Flavio Copes's write-up of Cloudflare's figures, across 14 URLs and a median of five runs against a warm Chromium pool, Kitesurf used 57.8 MiB for a screenshot against Chromium's 271 MiB.
It was also slower: 1,148 milliseconds for a screenshot against Chromium's 637, and 820 against 472 for HTML extraction. Copes writes that Kitesurf is in beta, free during the beta, and cannot yet complete bot-challenge handshakes that depend on real TLS fingerprints.
Less memory in exchange for more wall time is the tradeoff Cloudflare reports. Lightpanda reports less memory and less time, on its own crawl.
For related coverage of automated browsing, see the report on cfsolve, a Turnstile solver that needs no browser and the one on Firezone's sans-IO WireGuard library.
The unfinished piece is the 92 failing CORS subtests, which Bouvier's post says are still in progress.
Sources
More in Software
- 01Python 3.15 Ships a Built-In sentinel After Five Years of PEP 661The new type replaces the object() idiom with a one-line constructor that survives copying, pickling and type checking, but a Python 3.15 floor is the price of using it.
- 02Claude-Mem Hits 107,900 Stars, and Its Own Pages Disagree on LicenceThe Claude Code memory plugin has gained about 35,500 stars since May, but a May write-up says AGPL-3.0, the repository now says Apache-2.0, and the README describes an optional cloud sync that earlier coverage did not.
- 03Jepsen Found 96 Issues in ParticleDB 1.1.4 and Tells Users to Halt DeploymentKyle Kingsbury’s 31 October report, funded by vendor Orkes, finds lost writes and ignored constraints in healthy single-node runs.
- 04kube32 Runs Kubernetes Pods on ESP32-S3 Boards, and Its README Lists TLS Its Own Quick Start SkipsThe Apache 2.0 project, at 25 stars, runs WebAssembly pods under kubectl, but its SECURITY.md says the API serves plain HTTP with authentication off by default.