Search for types of proxies and you will find a dozen lists that do not agree with each other. One says there are three kinds, another says seven, a third puts SOCKS5 and residential side by side as though you had to choose between them. They are not contradicting each other so much as answering different questions without saying which one they picked.
Here is the thing that makes all of it fall into place. Type is four independent questions, not one, and every real proxy answers all four at once.
- Whose network does the address belong to? Residential, datacenter, ISP, mobile.
- Does the address change? Rotating or static.
- Who else is using it? Shared or dedicated.
- What protocol do you speak to it? HTTP or SOCKS5.
A rotating shared residential SOCKS5 proxy is a complete description. So is a static dedicated ISP HTTP proxy. When a provider advertises only one axis, the other three still have values, and the one they left out is frequently the one that decides whether the thing works for you.
Question one: whose network is it
This is the axis that does the most work, because it is the one destinations act on first and most cheaply.
Every public address belongs to an autonomous system, and which organisation owns which ranges is public information. A site can therefore classify you as hosting or household in one lookup, before it reads your request, before your behaviour matters, before anything you control comes into play.
Datacenter proxies run on machines in racks. They are fast, always on, cheap and abundant, and permanently identifiable as infrastructure. Rotating within hosting space does not help, because the next address resolves to the same kind of owner.
Residential proxies route through addresses that consumer ISPs assigned to home connections. They cannot be dismissed as a category, because that category is where the real customers live. You pay for that in speed, in price, and in the fact that you are borrowing something temporary.
ISP proxies are the hybrid: a machine in a data centre holding an address registered to a consumer provider. Datacenter behaviour, residential paperwork. The catch is supply, since each one depends on an arrangement made country by country, so coverage is genuinely thin rather than thin by choice.
Mobile proxies use carrier-assigned addresses. They sit at the top of the trust hierarchy for a reason that is arithmetic rather than flattery: carrier-grade NAT puts many subscribers behind one public address, so blocking it risks locking out real customers the site cannot distinguish from you. You are standing in a crowd, and that is the product.
| Type | Trust | Speed | Cost | Billed by |
|---|---|---|---|---|
| Datacenter | Lowest | Fastest | Lowest | Address |
| ISP | High | Fast | Mid | Address |
| Residential | High | Slower | High | Traffic |
| Mobile | Highest | Most variable | Highest | Access |
How the four families actually differ
Notice that billing changes with the family, and it is not a pricing decision. Residential is sold by the gigabyte because nobody can honestly rent you a specific household address by the month: the person on the other end may reboot their router or change provider. Datacenter and ISP are sold per address because the supply is a machine that stays put.
Question two: does the address change
Rotating means you connect to one gateway proxy and it assigns a different exit per request from a pool. Static means one address stays yours.
The reason rotation exists is arithmetic, not cleverness. Rate limiting is almost always counted per address, so ten thousand requests from one IP trips a threshold that the same ten thousand spread over a pool never approaches. You have not evaded anything. You stopped concentrating traffic in one place, which is also what a population of real visitors looks like.
The cost is that continuity becomes impossible. A login, a cart, a paginated crawl following a server-side cursor: all of them assume the same visitor returns. Rotate mid-flow and you have described a stolen session to whatever is watching, and it usually fails by silently logging you out rather than by erroring, which is far harder to diagnose.
A sticky session is the deliberate compromise: hold one exit for a defined window, then release it. On our network that is a field on the credentials rather than a different product, and the ceiling is 24 hours.
Question three: who else is on it
Shared splits an address between several customers. Dedicated assigns it to you alone.
What is being shared is not bandwidth, it is reputation. An address is only as trusted as everything ever done through it, and on a shared IP that includes strangers you cannot see or influence. The failure mode is worth picturing because it does not look like a failure: your requests still connect, latency is normal, nothing errors, and you simply start getting challenge pages or subtly wrong data on a target that worked yesterday.
Free public proxies are this taken to its limit, and our own numbers show the shape of it. Of 589,918 free proxies we have discovered and tested, 4,023 were alive at last measurement, and most of the rest never returned a single successful check in their entire recorded history.
The premium for dedicated is therefore not a premium on hardware. It is a premium on diagnosability: when something goes wrong on an address only you use, the cause is your own traffic, and a cause you can locate is a cause you can fix.
Question four: which protocol
HTTP proxies understand web requests. They can cache, route on what they see, and rewrite headers, which is also why they are the only family where anonymity grades are a question at all: an X-Forwarded-For header can hand the destination the very address the proxy was meant to hide.
SOCKS5 proxies are indifferent relays. They move whatever bytes you hand them without parsing, so they carry email, database connections, game protocols and UDP, and they never face the header question because nothing in the protocol was going to write one. They can also resolve hostnames at the exit, which is the fix for a DNS leak.
For ordinary web traffic, HTTP is the simpler answer and every tool supports it. For anything that is not HTTP, SOCKS5 is not the better choice, it is the only one.
So which do you actually need
Work backwards from the target, not forwards from a preference.
Does the site screen by network type? Test it. A great deal of scraping runs perfectly on datacenter addresses, because plenty of sites only count requests and never look at who owns the range. Paying residential rates for one of those is spending money on trust nobody asked you for. If it does screen, you need residential, ISP or mobile, and no amount of tuning rescues a datacenter address.
Does your task hold a session? If yes, you want static, or a sticky session long enough to finish. If no, rotating is what keeps every address under the limits.
Does a clean history matter? If a block would cost you a result rather than a retry, pay for dedicated. If not, shared is a sensible saving rather than a false economy.
Is your traffic HTTP? Then either protocol works and HTTP is simpler. If it is not, SOCKS5.
The mistake we see most often is not picking the wrong type. It is picking the most expensive type, finding it still does not work, and concluding the proxy was bad. Very often the address was fine and the request was announcing itself somewhere else entirely: a TLS fingerprint that belongs to a scripting library, a browser fingerprint reporting a time zone that contradicts the exit's country, or pacing no person could produce. Those layers sit above the network, and no address reaches them.
Our own residential is $0.44/GB, pay as you go, no subscription and a balance that does not expire, and we would rather tell you that datacenter is enough for your target than sell you the expensive answer to a question you did not have.