Choosing Mobile Proxies for Web Scraping

What scraping asks of a proxy

A scraper puts different demands on a proxy than an account session does. It sends many similar requests, cares more about throughput than identity, and is judged by how many pages come back complete.

Four things matter most. Exits in parallel: how many different addresses you can use at the same moment. Rotation: whether the exit changes per request, on a schedule or when you decide. Cost per page: what you pay for each page that arrives intact, including retries and blocked responses. Network type: whether the target treats carrier traffic differently from broadband or hosting traffic.

Mobile exits help on targets that challenge datacenter and some residential ranges heavily, because carrier addresses are shared by many ordinary subscribers through carrier-grade NAT. They do not make a scraper invisible. Request pace, headers, browser fingerprint and the order of page loads all still count.

Collect responsibly. The Robots Exclusion Protocol, standardized in RFC 9309, is how site owners state which paths crawlers may fetch. Respect it and the site's terms, keep request rates reasonable and avoid personal data you have no basis to collect. For collection techniques, see our web scraping proxy guide.

Five criteria before price

Billing model. Metered plans charge per GB and often add a monthly minimum. Dedicated devices charge per day, week or month. Light, uneven work suits metered; heavy, steady work suits per-device pricing.

Parallel exits. A shared network can give you hundreds of exits at once. A dedicated device gives one at a time. Count how many simultaneous addresses your crawler needs at peak.

Rotation control. Shared networks usually rotate per request or hold a sticky session set by a parameter. Decodo documents sticky sessions from 1 to 1,440 minutes with a 10-minute default (Decodo sticky sessions), and Oxylabs uses a session ID in the username to hold one IP across requests (Oxylabs session control). A dedicated line holds its address until you rotate it.

Targeting. Check whether you need country, city, ZIP, carrier or ASN, and whether the provider charges extra for it.

API and tooling. A crawler at scale needs programmatic rotation, health checks and replacement. Check what the provider's API exposes before you build around it.

Providers compared

Sources for each row: Decodo mobile pricing, Oxylabs mobile pricing, SOAX pricing, DataImpulse mobile proxies, ProxyEmpire pricing and IPRoyal mobile proxies. The Coronium row reads the live catalog when this page loads. Discount labels and plan names were as shown on the day; recheck at checkout. Our broader mobile proxy provider comparison covers more providers.

Reading the table fairly

Headline prices across these pages are not like for like, so adjust them before comparing.

Discount labels. Two providers showed prices as a percentage off a higher listed rate on the day we checked. Treat the discounted figure as current but not permanent, and check what renews at full price.

Minimums and plan fees. A low per-GB rate attached to a large monthly plan only applies if you buy that plan. For small crawls, the entry rate is the honest comparison.

Expiry and rollover. SOAX plan credits are valid for 60 days, DataImpulse traffic does not expire and ProxyEmpire rolls unused GB forward. If your crawl volume swings month to month, this changes the real cost more than the headline rate.

Caps on devices. Per-device pricing is not always unmetered. IPRoyal lists a 30 GB daily limit per dedicated proxy; Coronium lines are not metered except in Brazil, where they include 20 GB a month.

Tax. Several prices were shown before VAT. Add it if it applies to you.

What stops scrapers, and what a mobile exit changes

It helps to be precise about which defenses a mobile exit affects, because buying the wrong fix is the most common waste in scraping budgets.

Network reputation and type. Many sites score or block whole ranges: hosting providers first, then ranges with a history of abuse. A carrier exit changes this signal, since the address is shared with ordinary subscribers and reads as cellular in commercial databases. This is the main thing a mobile proxy buys you.

Rate limits. Sites count requests per address, per session, per account or per fingerprint. A new address resets only the first. If your crawler is blocked for pace, rotation alone will not hold; slow down and spread the work.

Browser and TLS fingerprints. A headless browser with default settings, or an HTTP client whose TLS handshake does not match the browser it claims to be, is visible on any network. Fix the client; the proxy cannot.

Behavior. Loading only data endpoints, never assets, at perfectly regular intervals looks automated however clean the address is. Vary timing and follow the navigation a real visitor would.

Challenges. When a challenge page appears, record it as a failure, back off, and rotate only after the pace has dropped. Rotating at full speed tends to burn through addresses.

A mobile exit is worth paying for when the first item is your bottleneck. If the others are, spend the effort on the crawler first and test again.

Page weight decides a metered bill

On per-GB plans, what you fetch matters as much as how many pages you fetch. An API response or bare HTML page can be a small fraction of a megabyte; a full browser render that loads scripts, images and fonts can run to several megabytes.

The weights are illustrative; measure your own by logging response sizes for a sample of pages. The line column assumes one line can carry the volume at your crawl pace. A crawler that needs many simultaneous exits needs several lines, which multiplies that column.

Two habits cut metered cost sharply: block images, fonts and media in headless browsers when you only need text, and fetch JSON endpoints a page already calls instead of rendering the page. Both also reduce load on the target.

Pick by workload

Mixing is normal. Many teams run the wide, light part of a crawl on metered traffic and the heavy or logged-in part on lines. Revisit the split each quarter, because crawls tend to grow in one direction: either toward more locations, which favors metered traffic, or toward more depth in the same markets, which favors lines. For the dedicated side, compare countries and carriers on the mobile proxies overview, and see web scraping with mobile proxies for setup patterns.

Testing on a paid 24-hour line

A day of real traffic tells you more than any table. Here is a test plan that works for any provider.

1. Fix the target and the crawler. Use the same pages, headers, browser profile and pace for every option, and change only the proxy.
2. Run for a full day. Targets behave differently at night and in the local evening.
3. Log four numbers per request: status code, response size, time to complete and whether the page parsed correctly.
4. Count only good pages. A 200 response that contains a challenge page is a failure.
5. Compute cost per 1,000 good pages. For a line: the period price divided by good pages, times 1,000. For metered traffic: GB used times the rate, divided by good pages, times 1,000.

Here is the skeleton of a crawl loop that rotates a dedicated line between batches rather than mid-batch, using the restart endpoint from Coronium's API:

import time
import requests

API = "https://api.coronium.io/api/v3"
HEADERS = {"Authorization": "Bearer YOUR_API_TOKEN"}
MODEM_ID = "YOUR_LINE_ID"
PROXY = "http://USER:PASS@HOST:PORT"
proxies = {"http": PROXY, "https": PROXY}

def crawl(urls):
    for url in urls:
        r = requests.get(url, proxies=proxies, timeout=30)
        yield url, r.status_code, len(r.content)
        time.sleep(2)                        # pace between pages

def rotate():
    r = requests.post(f"{API}/modems/{MODEM_ID}/restart", headers=HEADERS, timeout=150)
    return r.json().get("rotated")           # read the field, not the status code

for batch in batches:                        # your list of URL lists
    for row in crawl(batch):
        print(row)
    rotate()
    time.sleep(60)                           # never faster than the 60-second minimum

The restart call reports whether the address really changed in its rotated field. If it comes back false, wait and try again rather than starting the next batch on the old address.

On Coronium, the 24-hour plan is the test: buy one line on the carrier you want (the cheapest in-stock 24-hour line today: From $7), and run the plan above. If the result holds, move to 7 or 30 days, and ask for a volume code at 10 or more lines.

How we checked

Every competitor figure comes from that provider's own public pricing or documentation page, read on 30 October 2026, and is quoted as listed, including discount labels. Coronium figures come from the live catalog when this page renders. Coronium sells dedicated mobile lines and is listed here alongside competitors. We did not run a cross-provider scraping benchmark; the test plan above is how we suggest you run one.

Frequently asked questions