Best Headless Browsers for Web Scraping

The 2 MB page is an example: measure yours in the browser's network tab. A rendered page with images blocked is often a fraction of that.

The break-even, in one line. Your own cost per 1,000 pages is your server cost for the time those pages take, plus the traffic: up to $5.90 at 2 MB per page with ProxyField residential proxies. A hosted cost per 1,000 pages is the browser time those pages take at the provider's rate, plus its proxy charge. Put both on your real page weight and time per page; the cheaper one at your normal daily volume wins, and the hosted one wins at peaks your servers cannot hold.

How hosted browsers bill:

Hosted makes sense when:

1. Your peak needs many times the browsers one server runs (dozens or hundreds at once), and the rest of the day needs few.
2. Your code runs in a serverless or locked-down environment where you cannot launch Chromium.
3. Nobody on the team can own browser updates, crashes and memory limits.
4. You need results this week, before you could build a browser fleet.

Your own browsers win when volume is steady, you want full control over each request, and you want a cost per page you can predict: servers plus GB of traffic.

No browser to drive at all: ProxyField's Web Scraping API is designed to return a JavaScript-rendered page for $2.00 per 1,000 successful results, or $1.00 from 10,000 results in one purchase, billed on successful results only. It is in development: Contact sales.

A headless browser has two parts: the browser engine that renders the page and the driver library your code uses to control it. The table ranks the engines and modes for scraping; the next section compares the drivers. Facts checked against Playwright's browsers page and Puppeteer's supported browsers on October 27, 2026; see Sources below.

Not on the list: PhantomJS and other engines that are no longer developed. Old engines render modern sites badly and stand out to the sites you visit.

Head-to-head details: Puppeteer vs Playwright. Selenium for scraping: web scraping with Selenium and Python.

• Python: headless Chromium through Playwright for Python; Selenium if your team already uses it. Pyppeteer, the old Python port of Puppeteer, is unmaintained.
• Node.js: headless Chrome through Playwright or Puppeteer; Crawlee if you want crawling queues and browser pools built in.
• Java: Playwright for Java or Selenium; HtmlUnit is a Java-only headless browser without a real rendering engine, fine for simple pages. More in Java web scraping.
• C# and .NET: Playwright for .NET or Selenium; PuppeteerSharp is a community port of Puppeteer. See C# web scraping.
• Go: chromedp drives Chrome over the DevTools Protocol, and rod is another Go driver. See Go web scraping.

The cost comparison and the four triggers are at the top of this page. A hosted browser API, such as Browserbase or Browserless, runs the browsers on the provider's servers. Your code connects over a WebSocket and drives the remote browser with the same Playwright or Puppeteer calls; the provider handles scaling, browser updates and often the IPs. Some providers sell this as a "scraping browser", which adds their own proxy network and retries.

// Playwright: drive a remote Chromium over CDP (the endpoint comes from the provider)
import { chromium } from 'playwright';
const browser = await chromium.connectOverCDP('wss://REMOTE-BROWSER-ENDPOINT');
const page = await browser.newPage();
await page.goto('https://example.com');

// Puppeteer: the same with puppeteer.connect({ browserWSEndpoint: 'wss://...' })

Where the provider accepts external proxies, as Browserless does from its Prototyping plan, you can keep your ProxyField traffic: the remote browser exits through gate.proxyfield.io:7000 with your username and password, so the price per GB and the targeting stay the same whichever browsers run the pages. The way to pass the proxy differs by provider; see its docs.

If you would rather send a URL and get the rendered HTML back, without driving a browser at all, ProxyField's Web Scraping API is designed to fetch any URL as plain HTTP ($0.50 → $0.20 per 1,000 successful results) or with JavaScript rendering ($2.00 → $1.00 per 1,000), billed on successful results only; the volume price applies from 10,000 results in one purchase. The API is in development: Contact sales.

• Reuse browsers. Launch once and open new contexts or pages; launching a browser per URL is the most common waste.
• Cap concurrency per machine. Each browser needs CPU and a few hundred megabytes of memory; measure how many your server runs before pages slow down.
• Block what you do not need. Images, fonts, media and third-party trackers are often most of a page's bytes. Blocking them speeds pages up and cuts proxy traffic.
• Render only where needed. Fetch list pages with HTTP when possible and open a browser only for the detail pages that need it.
• Watch the results. Alert on empty fields or falling item counts, which signal a layout change or a block.

More techniques: headless browser scraping and how to make web scraping faster.

• Playwright, "Browsers": "Using the default Playwright configuration with the latest Chromium is a good idea most of the time"; headless runs use a separate Chromium headless shell by default; "You can opt into the new headless mode by using 'chromium' channel"; npx playwright install --with-deps chromium installs the browser and its system dependencies, --only-shell installs only the headless shell (for CI), --no-shell skips it, and npx playwright install --list lists installed browsers. Playwright supports Chromium N+1, which reaches Google Chrome and Microsoft Edge a few weeks later.
• Hosted browser pricing pages, Browserbase and Browserless: billing units, concurrency per plan and how proxies are billed, as in the table above.
• Puppeteer, "Supported browsers": Puppeteer 25.16.0 pairs with Chrome for Testing 156.0.8078.47 and Firefox 158.0.1; Puppeteer has used Chrome for Testing since v20.0.0 and stable Firefox since v23.0.0; the old headless mode is the separate chrome-headless-shell, selected with headless: 'shell'.