IPv6 Proxies Are Cheaper, But Can You Actually Use Them

Run a DNS lookup for AAAA records across your actual target list. Whatever fraction comes back with an IPv6 address is the ceiling on how useful an IPv6 proxy can ever be to you, because a client with only an IPv6 address cannot reach a server that only has an IPv4 one. If that fraction is low, no discount saves the purchase, because you will need an IPv4 plan for the remainder and you will be paying twice.

Then look at what you are actually renting. IPv6 is handed out in prefixes, not in single addresses, and reputation systems block prefixes. A vendor advertising ten million IPv6 addresses may be advertising one /64, which any competent anti-bot system treats as one entity. Price per prefix, never per address.

Where does that leave SparkProxy? Our public plans are IPv4 datacenter proxies, 1M+ addresses across 80+ countries, priced by concurrent threads from $75/mo. We do not sell IPv6 pools, so this article has nothing to sell you and no reason to soften the conclusion.

Everything starts here. An IPv6-only client reaches an IPv6-capable server, and nothing else. If your target publishes only an A record, an IPv6 proxy cannot deliver your request to it without a translation layer in between, and that translation layer changes what the target sees in ways covered further down.

The check takes one command per host:

# Does this target have an IPv6 address at all?
dig +short AAAA example.com
dig +short A example.com

Run it across your whole target list and count:

while read -r host; do
  aaaa=$(dig +short AAAA "$host" | head -n1)
  if [ -n "$aaaa" ]; then
    echo "AAAA  $host  $aaaa"
  else
    echo "none  $host"
  fi
done < targets.txt | tee aaaa-audit.txt

echo "coverage:"
awk '{print $1}' aaaa-audit.txt | sort | uniq -c

If dig is not available, the same answer comes from Python's resolver:

import socket

def has_ipv6(host):
    try:
        socket.getaddrinfo(host, 443, socket.AF_INET6)
        return True
    except socket.gaierror:
        return False

with open("targets.txt") as fh:
    hosts = [line.strip() for line in fh if line.strip()]

covered = [h for h in hosts if has_ipv6(h)]
print(f"{len(covered)}/{len(hosts)} targets publish AAAA "
      f"({len(covered) / len(hosts):.0%})")

Two warnings about interpreting the result.

An AAAA record means the host is reachable over IPv6. It does not mean the host treats IPv6 clients identically. Some sites serve IPv6 and apply stricter rate limits to it, or route it through a different edge with different rules, precisely because IPv6 is less commonly used by the traffic they want and more commonly used by traffic they do not. Reachability is necessary, not sufficient.

And check the hosts your target actually loads content from, not just the front door. A site whose HTML is served over IPv6 but whose product API sits on an IPv4-only subdomain has given you a half-usable target, which in practice means an unusable one.

There is a lot of published commentary about global IPv6 adoption percentages. Google and APNIC Labs both maintain continuously updated measurements, and both render the current figure as an interactive chart rather than as text, so no percentage is quoted in this article. Go and read them directly if you want the macro picture. For a buying decision they are close to irrelevant anyway, because those statistics measure end users with IPv6 connectivity, and you are not an end user. You are a client trying to reach servers, and server-side AAAA coverage on your specific list is the only number that governs your outcome.

This is the part that makes cheap IPv6 pools look cheap, and it is the single most important thing to understand before buying one.

IPv4 addresses are scarce and are allocated individually or in small blocks, which is why a single IPv4 proxy costs real money. IPv6 addresses are so abundant that allocation happens in prefixes: regional registries hand ISPs large blocks, and an end site typically receives a /48 or /56, while a /64 is the standard size of one subnet and contains 18 quintillion addresses.

So when a vendor advertises millions of IPv6 proxies, ask what prefix they came from. If the answer is one /64, then the honest description of the product is "one subnet, sliced arbitrarily fine". Every address in it shares a prefix, and reputation systems score prefixes. Our piece on subnet proxies makes the general argument; on IPv6 it is not a subtlety, it is the whole product.

The practical consequence is brutal and simple. A target that decides to block you does not block one /128 address. It blocks the /64, or more commonly the /48, and your ten million addresses become zero addresses in a single rule. On IPv4 the equivalent action costs the target something, because a /24 contains only 256 addresses and blocking it may catch legitimate users. On IPv6 a /64 belongs to exactly one customer by design, so blocking it is free of collateral damage and anti-bot vendors know it.

Which gives you the metric that matters:

useful diversity = number of distinct /48 prefixes in the pool
                   (NOT the advertised address count)

Ask for that number. A vendor who can answer it is selling something. A vendor who deflects to the address count is selling the number 18,446,744,073,709,551,616 at a discount, which sounds impressive and buys you one blocking decision. The same scepticism applies here as in how to spot a fake proxy provider before you buy and proxy pool size claims explained.

Some products sold as IPv6 proxies are IPv6 on the side facing you and IPv4 on the side facing the target, via NAT64, DNS64 or a plain protocol-translating gateway. That is not necessarily dishonest, and for some use cases it is genuinely convenient. It is also a completely different product from what most buyers think they are purchasing.

If the exit is translated, the target sees an IPv4 address belonging to the translation gateway. Which means the address diversity you were sold on the IPv6 side does not exist on the side that matters. Many IPv6 client addresses can share very few IPv4 exit addresses, and those exit addresses are shared with every other customer using the same gateway. You have bought a shared IPv4 proxy with an IPv6 front door, at IPv6 prices, and if you are lucky the price reflects that.

The question to ask, in writing, is exactly this: when my request reaches the target, is the source address an IPv6 address from the prefix I rented, or an IPv4 address belonging to your gateway?

The test is even simpler than the question. Send a request through the proxy to any service that echoes the client address and look at what comes back. If it is a dotted quad, the exit is IPv4 regardless of what the product is called.

curl -s -x "http://USER:PASS@your-vendor.example:PORT" https://api64.ipify.org

Now the arithmetic that decides the purchase. It is short.

Let X be the IPv6 price per unit of work, Y the IPv4 price for the same work, and c your measured AAAA coverage, expressed as a fraction between 0 and 1. Only the c portion of your targets is reachable, so your effective IPv6 cost per successful unit is X divided by c.

IPv6 is cheaper per successful request only when:

    X / c  <  Y

which rearranges to:

    X / Y  <  c

In words: the price ratio has to be lower than your AAAA coverage rate.

Work an example. If your audit says 30% of your targets publish AAAA, IPv6 has to cost less than 30% of what IPv4 costs before it breaks even, and that is before you account for the fact that the other 70% still needs an IPv4 plan. Two plans, two integrations, two sets of credentials, and a router in your code deciding which to use per host.

Published IPv6 proxy pricing is genuinely much lower than IPv4 pricing, which is why this test is worth running rather than dismissing. We are not quoting a specific vendor rate here, because the IPv6 pricing pages we could load today did not render figures we would be willing to print, and a price you cannot verify is not a price. Get a written quote, put X and Y into the inequality above, and let the arithmetic answer rather than the marketing.

One adjustment before you trust the result. The inequality assumes an IPv6 request that reaches an AAAA-publishing target succeeds at the same rate as an IPv4 one. If your test shows a lower success rate on IPv6, and it often will for the prefix reasons above, fold that into c as well: multiply your AAAA coverage by your measured relative success rate. The bar rises accordingly.

There are real cases, and they are more specific than the marketing suggests.

Targets you control. Load testing your own infrastructure, validating your own IPv6 deployment, testing how your application behaves for IPv6 clients. Here coverage is 100% by definition and cost is the only variable, so IPv6 wins outright.

Verifying IPv6 delivery. If you run ads, a CDN or a DNS-based service and need to confirm what an IPv6 client sees, an IPv4 proxy cannot answer that question at any price.

High-volume work against confirmed IPv6-friendly targets. Some sites are fully dual stack, apply no differential treatment, and do not block by prefix. Verify all three, then enjoy the price.

Regional work where the local access network is IPv6-first. Some markets are considerably further along than others, and testing from a client that looks like a local user means an IPv6 client.

What those cases share is that coverage is known to be high and the target is not adversarial. That is the profile. When the target is adversarial, the prefix problem dominates everything else and the discount stops being a discount.

The whole pool dies at once. Covered above. This is the defining failure mode, and it is not gradual. One rule, all addresses.

Dual-stack fallback hides the failure. A client configured with both address families will often retry over IPv4 when the IPv6 path fails, which means your job appears to work while not actually using the product you bought. If your IPv6 proxy costs less and you never see an error, verify the exit address rather than assuming.

Geolocation is worse. IPv6 geolocation databases are generally less mature than their IPv4 equivalents, and a prefix may be mapped to the allocation holder's registered location rather than to where the traffic actually leaves. If your work depends on appearing to be in a specific city, verify it per prefix rather than trusting the vendor's country label.

Some targets treat IPv6 as a signal in itself. A residential consumer with IPv6 is ordinary. A datacenter IPv6 prefix making thousands of requests is a very legible pattern, and it lives in a hosting ASN that is easy to identify. What is a datacenter ASN covers why that matters on either address family.

Tooling gaps. Bracket syntax in URLs, libraries that default to AF_INET, logging that truncates addresses, allowlists in third-party services that only accept IPv4 notation. None of these is hard, all of them cost an afternoon each, and they are invisible until you are mid-migration.

Send these in one message and judge the reply on specificity as much as on content.

1. How many distinct /48 prefixes does the pool span, and how many /64s?
2. Is the exit native IPv6, or translated to IPv4 at your gateway?
3. If translated, how many IPv4 exit addresses are there and are they shared across customers?
4. Which prefixes are they, so I can check their allocation and geolocation records myself?
5. What happens when a target blocks one of your prefixes? Do I get a different one, and at what cost?
6. Is authentication by user and password, by IP allowlist, or both, and does the allowlist accept an IPv6 client address?
7. What is the minimum term, and can I test on my own targets before committing to it?

Question 5 is the one that separates serious vendors. Prefix-level blocking is not an edge case on IPv6, it is the expected outcome, and a vendor with no answer for it has not run this product at scale against defended targets.

Fifteen minutes: the AAAA audit. Run the loop from the DNS check section across your full target list. Write the coverage percentage down. This is c, and it is the number the entire decision hangs on.

Ten minutes: verify the exit. Buy the smallest possible quantity and confirm what a target actually sees. If the echo service returns an IPv4 address, stop and re-read the product description.

Twenty minutes: measure relative success. Take twenty targets that do publish AAAA, hit each one fifty times over IPv6 and fifty times over an IPv4 proxy, and compare success rates. The procedure in how to test proxies applies unchanged. Differences here are usually large and obvious rather than marginal.

Fifteen minutes: run the inequality. Put your quoted X, your IPv4 Y, and your coverage times relative success rate into X / Y < c and see whether it holds. If it does not hold, the discount is not a discount.

If all you actually need is for requests to arrive successfully and you do not care which address family carries them, a managed API removes the question entirely:

curl -G "https://scrape.sparkproxy.io/api/v1" \
  -H "X-API-Key: YOUR_API_KEY" \
  --data-urlencode "url=https://example.com/catalog/page/3" \
  --data-urlencode "render_js=false" \
  --data-urlencode "premium_proxy=true" \
  --data-urlencode "country_code=US"

The exit selection is the vendor's problem rather than yours, plain fetches cost 1 credit, and the free tier is 1,000 credits with no card, which is enough to compare against any IPv6 quote you have been given.

For most commercial data collection, IPv6 proxies are not worth buying today, and the reason is structural rather than temporary. The discount exists because IPv6 addresses are abundant, and the same abundance is what makes prefix-level blocking cheap and effective for the sites you want to reach. You cannot have the low price without the property that causes the low price.

Buy them when coverage is known, the target is not adversarial, and the exit is verified native. That combination is real and the savings in it are real. Assume none of those three things without checking, and the cheapest proxies you have ever bought become the most expensive per successful request.

And check the arithmetic rather than the argument. X / Y < c takes two minutes once you have run the audit, and it will tell you more than any comparison article, including this one. For the wider pricing picture across proxy types, how much do proxies cost has the full ladder.