Atlas targets
One success rate is not a success rate.
A network that clears an undefended page can fail a search engine. So every network gets the same requests to the same 8 targets, and each result is published on its own.
- HTTP
- HTTPS
- Marketplace
- Social
- Search
- Challenge
- Profile
- AI
- Same requests
- Written pass rules
- Easy to hardest
- Results per target
Ranked by Atlas Score across 18 networks
Shifter94.1#1
Oxylabs83.4#2
Decodo81.5#3
FleetProxy80.6#4
Geonode80.3#5
DataImpulse80.1#6
ProxyEmpire79.9#7
Webshare79.3#8
NodeMaven77.7#9
SOAX77.5#10
Proxy-Seller77.2#11
Byteful75.6#12
Infatica74.6#13
Massive74.2#14
Evomi73.1#15
Rayobyte71.5#16
Proxy-Cheap70.5#17
IPRoyal66.4#18
Shifter94.1#1
Oxylabs83.4#2
Decodo81.5#3
FleetProxy80.6#4
Geonode80.3#5
DataImpulse80.1#6
ProxyEmpire79.9#7
Webshare79.3#8
NodeMaven77.7#9
SOAX77.5#10
Proxy-Seller77.2#11
Byteful75.6#12
Infatica74.6#13
Massive74.2#14
Evomi73.1#15
Rayobyte71.5#16
Proxy-Cheap70.5#17
IPRoyal66.4#18Easiest to hardest
The same 8 targets for every network, from an undefended page to an AI assistant.
- Baseline
Generic HTTP
A plain page over HTTP, with no defence.
Networks tested
18 of 18
- Baseline
Generic HTTPS
The same plain page, over HTTPS.
Networks tested
18 of 18
- E-commerce
Marketplace
A product page on a large marketplace.
Networks tested
18 of 18
- Social
Social
Logged-out pages on two social networks, Instagram and X.
Networks tested
18 of 18
- Search
Google Search SERP
A search engine results page.
Networks tested
18 of 18
- Anti-bot
Managed challenge page
A site behind a managed bot challenge.
Networks tested
18 of 18
- Social
Public professional profile
A logged-out public profile page.
Networks tested
18 of 18
- AI
AI
The same question, asked of three AI assistants.
Networks tested
18 of 18
Read the exact rules
Every network is sent the same requests to every target, from the same locations, on the same schedule. A response only counts as a success when it meets the target's written pass rule: an HTTP 200 that carries a CAPTCHA or a block page is a failure. Each target is scored over the last 24 hours.
Generic HTTP
Difficulty 1 of 5Passes when: Response body is the exit address the target saw, and it parses as an IP.
The baseline. With nothing defending the page, a failure here is the proxy network, not the target.
Generic HTTPS
Difficulty 1 of 5Passes when: CONNECT tunnel established, certificate validates, and the body parses as an IP.
Adds the TLS handshake through the proxy, so it shows what encryption costs in failures and speed.
Marketplace
Difficulty 2 of 5Passes when: The product page loads with its title and price or buy box, not a captcha.
Price and stock pages are the most common scraping job, behind moderate bot defences.
Social
Difficulty 3 of 5Passes when: Never run itself: computed over the tries of its two sources, Instagram and X.
Social apps check the connection as well as the IP and limit logged-out visitors, so they catch pools that only look residential.
Google Search SERP
Difficulty 4 of 5Passes when: The results page shows organic results, not the /sorry captcha or a consent wall.
Search engines challenge shared and recycled IPs quickly, so they expose pools that are used too hard.
Managed challenge page
Difficulty 4 of 5Passes when: The managed challenge clears and the protected page shows its expected text.
A managed challenge scores the exit IP's reputation before any page is served.
Public professional profile
Difficulty 5 of 5Passes when: The public profile shows its name and headline, not the authwall or a 999 block.
Profile pages refuse known proxy ranges and hosting networks, which tests how residential a pool really is.
AI
Difficulty 5 of 5Passes when: Never run itself: computed over the tries of its three AI sources.
AI assistants ask doubtful visitors to sign in instead of answering, so they show which pools still pass as real people.
Every network, target by target
Success on each target over the last 24 hours, for every network in Atlas.
| Network | Average | HTTP | HTTPS | Marketplace | Social | Search | Challenge | Profile | AI | |
|---|---|---|---|---|---|---|---|---|---|---|
Shifter | 9.5 | 4.2 | 4.2 | 5.2 | 7.2 | 9.4 | 8.7 | 1.6 | 4.7 | |
Oxylabs | 7.9 | 8.9 | 8.9 | 8.1 | 10.0 | 1.0 | 9.4 | 5.8 | 4.2 | |
Decodo | 9.7 | 9.5 | 9.5 | 8.7 | 7.2 | 7.6 | 7.1 | 8.7 | 8.4 | |
FleetProxy | 7.1 | 3.6 | 3.6 | 9.4 | 4.1 | 5.5 | 10.0 | 6.3 | 1.0 | |
Geonode | 5.8 | 6.3 | 5.8 | 1.6 | 5.5 | 1.6 | 4.5 | 3.1 | 3.6 | |
DataImpulse | 8.6 | 7.4 | 7.4 | 4.9 | 10.0 | 2.1 | 4.5 | 2.3 | 6.0 | |
ProxyEmpire | 7.0 | 5.8 | 5.2 | 4.2 | 3.0 | 8.4 | 7.1 | 3.9 | 1.5 | |
Webshare | 10.0 | 8.4 | 6.3 | 4.2 | 10.0 | 3.9 | 6.3 | 3.9 | 10.0 | |
NodeMaven | 6.1 | 2.6 | 2.1 | 2.6 | 1.6 | 6.5 | 2.3 | 8.7 | 2.6 | |
SOAX | 8.8 | 4.7 | 8.4 | 2.3 | 5.5 | 4.9 | 8.7 | 6.8 | 8.9 | |
Proxy-Seller | 8.7 | 2.1 | 2.6 | 5.8 | 4.1 | 8.4 | 7.1 | 5.8 | 6.0 | |
Byteful | 9.3 | 7.9 | 6.8 | 10.0 | 10.0 | 3.1 | 4.5 | 8.7 | 7.9 | |
Infatica | 7.5 | 3.1 | 3.1 | 5.8 | 2.9 | 7.4 | 2.3 | 3.9 | 6.8 | |
Massive | 6.7 | 10.0 | 10.0 | 6.8 | 10.0 | 3.9 | 4.5 | 10.0 | 2.1 | |
Evomi | 9.4 | 6.8 | 7.9 | 7.4 | 10.0 | 10.0 | 2.3 | 5.2 | 7.4 | |
Rayobyte | 8.8 | 1.5 | 1.0 | 2.9 | 7.2 | 2.9 | 8.7 | 7.4 | 9.5 | |
Proxy-Cheap | 5.4 | 1.0 | 1.5 | 1.0 | 2.4 | 6.5 | 1.0 | 3.9 | 3.1 | |
IPRoyal | 7.5 | 5.2 | 4.7 | 3.6 | 6.8 | 2.3 | 7.1 | 1.0 | 5.2 |
Read the exact rules
Each cell is a score from 1 to 10 for how often that network's requests to the target met its written pass rule over the last 24 hours, against every other network tested there. Every request counts, and every time slot weighs the same, so a busy hour cannot drown out a quiet one.
Challenges and blocks count as failures
A CAPTCHA or interstitial usually arrives as HTTP 200. It still fails the pass rule, so it lowers the cell rather than hiding inside it.
Why only 24 hours
Targets change their defences often. A longer window would mix different defence states into one number, so per-target results stay on the last 24 hours.
Latency is measured on one target
Speed is measured on Generic HTTP only, because HTTPS adds handshake round trips that would blur the comparison. The other targets report success and failures.
Results older than a day
Browser targets come round about once a day, so a network may have no run on a target in the last 24 hours. Its tile then shows its latest result from the last 7 days, or the last 30, as the hover says. A tile is empty only when the network has never been tested on the target, and it is never counted as zero.
Plain HTTP latency, p50
- Shifter 531 ms
- Oxylabs 516 ms
- Decodo 541 ms
- FleetProxy 903 ms
- Geonode 824 ms
- DataImpulse 480 ms
- ProxyEmpire 722 ms
- Webshare 814 ms
- NodeMaven 537 ms
- SOAX 438 ms
- Proxy-Seller 776 ms
- Byteful 535 ms
- Infatica 751 ms
- Massive 353 ms
- Evomi 417 ms
- Rayobyte 969 ms
- Proxy-Cheap 1.09 s
- IPRoyal 1.08 s
Why the hard ones decide a purchase
An average over easy pages hides the targets most buyers actually pay proxies for.
-
A CAPTCHA arrives as a 200
Defended sites answer with a challenge page and a success code. Counting status codes would call that a success; the pass rule does not.
-
Busy pools get noticed
Search engines and social sites flag addresses that many customers share. Hard targets show how hard a pool is used.
-
Reputation is scored first
Managed challenges and queues judge the exit IP before serving anything. Only clean, consumer addresses get through.
How one request gets a verdict
Next · 04 Coming next Which targets are we adding next?Targets we want to add
The suite is versioned and open to proposals. A target has to earn its place.
- Its pass rule can be checked by code, so nobody has to judge a result.
- It can be requested within the site's terms and rate limits.
- Its result is not already predicted by a target in the suite.
-
Travel and fare search
Heavily defended, sensitive to location, and a large share of proxy spend.
-
App store and review pages
Defences built to tell real home connections from the rest.
-
Ad verification view
Needs a full page load, which shows how bandwidth is counted.
-
Streaming manifest
Spots datacenter ranges fast, and often decides a purchase.
-
Login wall
Shows whether an exit IP is trusted enough to be offered a session.
-
IPv6-only page
Almost no provider publishes honest IPv6 support.
Know a target buyers care about that is missing?
Suggest a targetTest Shifter on your own target
Shifter is measured on the same targets as every network in Atlas. Order with a 3-day money-back guarantee and send it to the site you care about.