IPv4 vs IPv6: Differences, Speed, Security and Adoption
IPv4 and IPv6 are two versions of the Internet Protocol, the addressing layer that lets devices send and receive data. IPv4 uses 32-bit addresses and its supply of new addresses ran out years ago. IPv6 uses 128-bit addresses and is now used by roughly half of users on some measures. This guide compares the two, tests common speed and security claims, shows how to check what your device and a website actually support, and explains when IPv6 proxies make sense for scraping.
IPv4 vs IPv6 at a glance
The clearest difference is address size. IPv4 addresses are 32 bits long and written as four decimal numbers separated by dots, such as 192.0.2.1. IPv6 addresses are 128 bits and written as eight hexadecimal groups separated by colons, such as 2001:db8::1. IPv6 also changes how packets are handled: it uses a fixed 40-byte header, removes the header checksum, drops broadcast in favor of multicast and anycast, and allows devices to configure their own addresses through router advertisements.
The two protocols run side by side in most networks. Google first recorded more than half of its users reaching it over native IPv6 on March 28, 2026, while other measures such as APNIC's worldwide capability figure were lower. The comparison table below summarizes the main technical differences.
What is IPv4?
IPv4 is the first widely deployed version of the Internet Protocol. It gives each connected device a 32-bit address, allowing about 4.3 billion addresses in total. When RFC 791 defined it in September 1981, that seemed like a generous supply. The growth of personal devices and always-on connections used it up.
The Internet Assigned Numbers Authority handed out the last five blocks of IPv4 addresses on February 3, 2011. Networks stretched the remaining supply with Network Address Translation, which lets a home router share one public address among many private devices, and with carrier-grade NAT, which shares a pool of public addresses among mobile subscribers. Those techniques keep IPv4 working, but they also break the original end-to-end model: a device behind NAT can make outgoing connections more easily than it can accept incoming ones.
What is IPv6?
IPv6 is the current version of the Internet Protocol, standardized in RFC 8200. Its 128-bit addresses allow about 340 undecillion addresses, which is enough to assign a very large number to every person on Earth and still have room left. It was designed to replace IPv4, but the two protocols now run together in most networks.
IPv6 notation takes some practice. An address can be written in full as eight hexadecimal groups, such as 2001:0db8:0000:0000:0000:0000:0000:0001. Leading zeros in each group can be dropped, and one run of consecutive zero groups can be replaced by a double colon. Applying both rules gives 2001:db8::1. RFC 5952 recommends a single compressed form so that the same address is written consistently. Most IPv6 addresses also split into a 64-bit network prefix and a 64-bit device identifier.
Address space and notation
IPv4's 32 bits allow 2^32 addresses, about 4.3 billion. IPv6's 128 bits allow 2^128, about 3.4 × 10³⁸, which is roughly 2^96 times as many. Another way to picture the gap: IPv6 could give every person on Earth about 10 quintillion copies of the entire IPv4 address space.
The written forms differ too. IPv4 addresses are always written in full as four numbers from 0 to 255. IPv6 addresses are usually compressed. That matters for logs, allowlists, and blocklists: the same IPv6 address can appear in more than one valid-looking form, so comparisons should use the standard compressed form. IPv6 addresses also need square brackets in URLs, as in https://[2001:db8::1]/, because the colons would otherwise be confused with a port separator.
Packet headers and fragmentation
An IPv4 header is 20 to 60 bytes depending on options and includes a checksum that every router recalculates. IPv6 uses a fixed 40-byte header, moves optional features into extension headers, and removes the header checksum, which saves routers work on every packet.
Fragmentation is another important difference. In IPv4, a router that encounters a link too small for a packet can split it, unless the Don't Fragment flag is set. In IPv6, only the sender may fragment. A router that cannot forward an oversized packet drops it and returns an ICMPv6 Packet Too Big message that includes the link's limit. Every IPv6 link must carry packets of at least 1,280 bytes. Because ICMPv6 is part of the control plane, blocking all ICMPv6 messages can break connections as soon as a full-size packet is sent.
Address configuration
IPv4 devices get addresses from a DHCP server or from manual configuration. IPv6 adds stateless address autoconfiguration, or SLAAC, defined in RFC 4862. A device listens for a router advertisement carrying the network's /64 prefix and generates the last 64 bits itself, with no server involved. DHCPv6 is the other option: it can assign addresses centrally, or run beside SLAAC to deliver settings like DNS servers, and ISPs rely on it to hand whole address blocks down to home routers.
Separately, each IPv6 interface auto-generates a link-local address in the fe80:: range. This address never leaves the local segment; routers and neighboring devices use it for on-link communication. Early SLAAC implementations built the device half of the address from the network card's hardware address, which made a laptop recognizable as it moved between networks. Temporary addresses, specified in RFC 8981, replace that with a random device half for new connections. They last one day, are retired after two days, and are now used by most desktop and mobile operating systems for outgoing traffic.
NAT and end-to-end addressing
IPv4 uses NAT and carrier-grade NAT to stretch its address supply, but those techniques complicate hosting a local server or using peer-to-peer applications. Such applications often need NAT traversal or relay servers to work around the problem.
IPv6 gives every device a public address. RFC 6177's guidance says homes and businesses should receive more than a single /64, such as a /56 with room for 256 subnets or a /48 with room for 65,536, though some ISPs still hand out less. A public address does not mean an open door: RFC 6092 recommends that IPv6 home routers block unsolicited inbound connections by default, and a firewall is still necessary.
Broadcast, multicast, and neighbor discovery
IPv4 uses broadcast to reach every device on a local network, most visibly in ARP. IPv6 replaces broadcast with multicast and anycast. ARP is replaced by Neighbor Discovery, defined in RFC 4861, which runs over ICMPv6 and also handles route discovery, address configuration, and duplicate address detection.
Watch a capture and the old ARP request shows up as an ICMPv6 Neighbor Solicitation addressed to a solicited-node multicast group in the ff02::1:ff range. The reply is a Neighbor Advertisement, and a device joining the network also emits Router Solicitations and receives Router Advertisements. Firewall rules should not block ICMPv6 wholesale: Neighbor Discovery uses ICMPv6 types 133 to 137, and dropping those messages breaks address resolution and autoconfiguration on the local network.
DNS records and dual stack
IPv4 addresses live in A records and IPv6 addresses in AAAA records, and the two sets are managed separately. A domain may publish one, the other, or both, and the bare domain and the www hostname need not match. A dual-stack device holds both address types and chooses one for each connection.
Networks that drop IPv4 still need a way to reach IPv4-only sites. NAT64, defined in RFC 6146, translates IPv6 packets to IPv4 at the network edge. DNS64, defined in RFC 6147, makes IPv4 invisible to devices by synthesizing AAAA records for names that only have A records. 464XLAT, defined in RFC 6877, adds a small translator on the device so applications that insist on IPv4 still work. IPv6-mostly networks can use DHCP option 108, defined in RFC 8925, to let capable devices decline an IPv4 address. In each case, the website at the far end sees an ordinary IPv4 connection from the translator's address.
Is IPv6 faster than IPv4?
Not by design. IPv6 has no built-in speed advantage that users would notice, and most measurements put the two protocols close together. When IPv6 is faster, the usual reason is that the path avoids carrier-grade NAT or has better routing. When it is slower, the cause is often a tunnel, poor peering, or a misconfigured network. That is why most IPv6 vs IPv4 speed benchmarks compare two network paths rather than the protocols themselves.
Happy Eyeballs, specified in RFC 8305, makes speed harder to measure. Most browsers and operating systems try IPv6 first and start an IPv4 connection in parallel if IPv6 does not connect within a fraction of a second. The system then uses whichever connection succeeds first. A slow or broken IPv6 path may therefore appear only as a brief delay that the user never notices. Simple scripts and scrapers often do not implement Happy Eyeballs, so a host with a broken IPv6 route can stall every request until the timeout expires before IPv4 is tried.
In gaming, NAT type matters more than the protocol. A strict NAT rating, common behind carrier-grade NAT, limits incoming connections. IPv6 can avoid that problem because each console gets a public address, but only for consoles and games that support IPv6.
Is IPv6 more secure than IPv4?
No, not by itself. The claim that IPv6 is safer because IPsec is mandatory is out of date. IPsec was required only in early specifications, became a should since RFC 6434 in 2011, and was never mandatory to use. Security depends on firewalls, configuration, and monitoring, which both protocols need.
Several common claims about IPv6 security are misleading. IPsec is not mandatory, and it works over IPv4 too. NAT blocks unsolicited incoming traffic only as a side effect of connection tracking; the IETF states that NAT was not developed as a security mechanism. A stateful firewall gives the same protection on IPv6. IPv6 cannot be scanned in the sense that brute-forcing a whole /64 is infeasible, but attackers can narrow the search with predictable addresses, DNS records, reverse-DNS lookups, and addresses that embed an IPv4 address or port number, as described in RFC 7707. A server with a public hostname is as easy to find as it is on IPv4.
Address tracking is another area where early behavior differs from current practice. Modern systems use random identifiers and rotating temporary addresses. The network prefix still identifies a household for as long as the ISP keeps it the same, much as a home's public IPv4 address does, and it can be more specific than an IPv4 address shared by many users through carrier-grade NAT. Microsoft advises against disabling IPv6 because some Windows components may stop working; its recommended alternative is to prefer IPv4 over IPv6 through prefix policies. A VPN or proxy that only handles IPv4 can leave IPv6 connections going out directly and revealing the real IPv6 address, as noted in RFC 7359.
How much of the internet uses IPv6?
About half of users can reach the internet over IPv6, yet only around a third of websites are reachable that way. Which figure you see depends on the method: Google tallies users who reach Google, APNIC tallies devices with IPv6 capability, and Cloudflare tallies requests hitting its sites. The table below shows where IPv6 adoption stood on the main measures at the dates given. These are dated reference figures, not a live reading, so check current sources before relying on them.
Why IPv4 is still everywhere
If the world ran out of IPv4 addresses more than 15 years ago, why do two-thirds of websites still work only over IPv4? Running out of new addresses did not make IPv4 unusable; it made addresses more expensive, but not expensive enough to force a direct swap. IANA handed its last blocks to the five regional internet registries on February 3, 2011, and the registries then exhausted their own pools at different times. APNIC moved to rationing in April 2011, ARIN ran out in September 2015, the RIPE NCC made its last allocation in November 2019, and LACNIC assigned its last available block in August 2020. AFRINIC, serving Africa, entered its final exhaustion phase in January 2020. Organizations that still need IPv4 generally queue on waiting lists for returned addresses or purchase them from holders with spare capacity.
Addresses sold for an average of $20.04 each in the first half of 2026, according to transaction data published on CircleID. The largest cloud providers pass the cost on: AWS, Google Cloud, and Azure each charge $0.005 an hour for a public IPv4 address. Even so, IPv4 does not cost enough to force a direct swap. Nearly every site still works over IPv4, so a network that adds IPv6 gains access to almost nothing it could not already reach, while a site that dropped IPv4 would lose roughly half of visitors whose connections have no IPv6. Dual stack lets the two protocols run side by side indefinitely, and NAT, carrier-grade NAT, and translation tools such as NAT64 cover the gaps for networks short of addresses.
IPv6 reachability across 1,050 sites
A study requested 1,050 sites over both protocols from two servers, five times each, and compared the results. About half of the most popular websites loaded over IPv6, but fewer than a third of the sites people most often scrape did. When a site treated the two protocols differently, it was more likely to block or rate-limit IPv4 than IPv6.
The main sample was the top 1,000 domains on the Tranco list, using list 8PXYV generated on October 9, 2026. Tranco ranks domains rather than websites, so each domain was tested at the bare domain and at www, and counted as reachable if either name worked. The second sample was 50 commonly scraped sites: 9 marketplaces, 9 retailers, 7 travel sites, 7 social platforms, 6 search engines, 6 job boards, and 6 real estate portals. For each site, the homepage and one page a scraper would want were tested.
The test ran from two DigitalOcean servers with native IPv6, one in Atlanta and one in Frankfurt. Each server looked up A and AAAA records with dig, then requested every URL with curl, once forcing IPv4 and once forcing IPv6, with the order randomized for each page. Each server made five full passes between 23:10 UTC on October 9 and 01:07 UTC on October 10, 2026. Status code, connection time, and time to first byte were recorded after DNS lookup, before any redirect. A second request followed redirects to record the final status and determine whether the page was a challenge or block page. A page counted as loading over a protocol if the final page arrived, with any status code, in at least three of the five passes from that server. Challenge and block pages were recognized by the signatures of major bot-protection services, generic access-denied titles, and HTTP 429 responses.
Of the 1,000 Tranco domains, 41.3% published an IPv6 address on either the bare domain or www. That figure is held down by content-delivery, advertising, and tracking domains that do not serve a website at all. Among the 725 domains that served a page over IPv4 from both servers, 54.1% published an IPv6 address and 49.1% loaded over IPv6 from both. For comparison, Common Crawl's July 2026 survey of the top million web hosts found 35.7% publishing an IPv6 address and 31.7% answering from all five of its test locations.
The scraping targets lagged behind. Only 15 of the 50 published an IPv6 address, including Google, Bing, Yahoo, Yandex, Amazon, IKEA, Expedia, Kayak, Hotels.com, Trivago, ZipRecruiter, and YouTube. eBay, Walmart, Etsy, Target, Booking.com, Airbnb, Zillow, and Indeed did not. Six targets failed over IPv4 from at least one server and were left out of the comparison. Of the remaining 44, 31.8% published an IPv6 address and 29.5% loaded over IPv6 from both servers.
Speed was effectively the same on both protocols. Medians in Atlanta were 1.3 ms to connect on both, and 49.8 ms on IPv4 versus 49.2 ms on IPv6 to the first byte. In Frankfurt, the medians were 1.2 ms versus 1.5 ms to connect and 108.9 ms versus 113.0 ms to the first byte. Of the 16 pages that treated the protocols differently, 8 blocked or rate-limited only IPv4 and 2 only IPv6. Where the bare domain and www disagreed, only www had an IPv6 address in 112 cases and only the bare domain in 20. The results and target list are published on GitHub.
How to check whether you're using IPv4 or IPv6
Most modern operating systems and internet providers switch between IPv4 and IPv6 quietly, so you may be using both without realizing it. The checks below show what your device has, what the internet sees, and what a given site supports.
On Linux, run ip -4 addr and ip -6 addr. On Windows, run ipconfig. On macOS, run ifconfig. Look for inet (Linux and macOS) or IPv4 Address (Windows) for IPv4, and inet6 or IPv6 Address for IPv6. A lone fe80:: address means IPv6 is switched on locally but the network is not handing out routable IPv6, since fe80:: only works on the local segment. A global address beginning with 2 or 3 confirms your network provides real IPv6.
To see your public addresses, run curl -4 https://ifconfig.co and curl -6 https://ifconfig.co on Linux or macOS. On Windows, use curl.exe -4 https://ifconfig.co and curl.exe -6 https://ifconfig.co. Each command returns the address remote sites observe for that protocol. If the IPv6 call fails with something like Couldn't connect to server rather than returning an address, your connection lacks a working IPv6 route. For a browser check, test-ipv6.com runs a series of IPv4 and IPv6 connections and gives your setup a score out of 10, flagging common problems such as IPv6 that is configured but broken.
To check whether a site supports IPv6, run dig AAAA example.com +short on Linux or macOS, or nslookup -type=AAAA example.com on Windows. One or more IPv6 addresses in the output means the site has an AAAA record. An empty result means the site is IPv4-only. An AAAA record alone does not guarantee the site serves content over IPv6, so confirm with an actual request: curl -6 -sS -o /dev/null -w "%{http_code} %{remote_ip}\n" https://example.com. A status code such as 200 followed by an IPv6 address means the site answered over IPv6. Your internet provider must support IPv6 for this command to work.
Forcing one protocol
With curl, -4 forces IPv4 and -6 forces IPv6. In Python, httpx can be pinned to one protocol by giving the connection a local bind address: 0.0.0.0 selects IPv4 and :: selects IPv6. The requests library has no direct setting for this, but you can override the function its underlying library, urllib3, uses to choose which kind of address to look up. That change applies to every request in the program, so set it once before making requests.
Forcing the protocol per client or per script is useful in scraping because you can choose the protocol for each job without changing anything else on your machine.
Prefer IPv4 without disabling IPv6
Microsoft advises against disabling IPv6 because some Windows components stop working without it. Turning it off also removes your access to IPv6-only services and to the IPv6 fallback that can rescue a failing IPv4 path. When IPv6 is causing trouble, the safer remedy is to make the system try IPv4 first while leaving IPv6 enabled.
On Windows, run the registry command in an administrator Command Prompt and restart. The value 0x20 is Microsoft's recommended prefer-IPv4-over-IPv6 setting. It changes only the order in which Windows tries addresses, so IPv6 still works when IPv4 is not available. On Linux, add the precedence lines to /etc/gai.conf. The final line lifts IPv4 above IPv6 in the sequence programs use when choosing an address. The first four lines must be present because any precedence entry causes the system to drop its entire default table, so the file has to spell it out again. The change applies to programs that rely on the system's address ordering, which includes most command-line tools and scripts.
When your network provides IPv6 and the earlier checks succeed, keep it enabled: you reach IPv6-only services and give up nothing on IPv4. Only set a preference if IPv6 is broken somewhere along your path. For scraping, force the protocol inside your code instead of reconfiguring the whole system.
IPv4 vs IPv6 proxies for web scraping
IPv6 proxies cost far less than IPv4 proxies, but a lower price does not make them harder to block. They can only reach sites that answer over IPv6, and many large sites treat an entire /64 range as a single client, so a thousand addresses drawn from one range may register as one visitor. Against hardened targets, IPv4 datacenter or residential proxies remain the more dependable option. A single /48 block of IPv6 contains 65,536 /64 subnets, each holding roughly 18 quintillion addresses, whereas an IPv4 address sold for an average of $20.04 in the first half of 2026. Proxy pricing reflects that gap: at the time of writing, Proxy6 listed IPv6 proxies at $0.08 to $0.26 per IP per month, against $1.24 to $1.50 for its IPv4 proxies, and Proxy-Seller advertised IPv6 from $0.02 per IP against $0.49 for IPv4. Rate limits and blocks rarely single out one IPv6 address. Cloudflare's documentation for its original rate-limiting product states that it triggers when an individual IPv4 address or IPv6 /64 IP range exceeds a rule threshold, which means all addresses within a single /64 are treated as one client. Since a /64 usually maps to one household or one server, range-based grouping is the obvious default for any site running bot protection. An IPv6 proxy is only useful if the target answers over IPv6. The test described above found that Google, YouTube, Wikipedia, and Amazon published an IPv6 address, loaded over IPv6, and returned the same response as IPv4. eBay did not publish an IPv6 address and did not load over IPv6. When a site publishes no AAAA record, an IPv6-only exit has no destination to reach. Depending on the provider, the request either fails outright or is silently routed through an IPv4 fallback, so you are no longer using the addresses you paid for and may be sharing a much smaller IPv4 pool with other customers. Use IPv6 proxies only when four things are true: the target supports IPv6, it serves the same content over IPv6 as over IPv4, it is lightly protected, and your provider spreads you across many /64s and /48s. If any of those fail, use IPv4 datacenter proxies for easy targets and residential proxies for hard ones.
A practical protocol checklist
Use this checklist to separate the three decisions people often mix together: what your device should do, what a website supports, and what a scraper should use.
- Device: run the local address commands. If you see a global IPv6 address starting with 2 or 3, your network provides IPv6. If you see only fe80::, IPv6 is enabled locally but not provided by the network.
- Path: run the curl -4 and curl -6 public-address checks. If the IPv6 command fails, your connection has no working IPv6 path. Do not disable IPv6 to fix a broken path; prefer IPv4 or fix the route.
- Site: look up the AAAA record, then make a real IPv6 request. A published address is a hint, not proof that the site serves content over IPv6.
- Scraper: if the target is IPv6-ready, lightly protected, and your proxy provider spreads you across many /64s, IPv6 proxies can be economical. If the target is hardened, IPv4 datacenter or residential proxies are more reliable.
- Failure distinction: a broken IPv6 path causes timeouts or fallback delays. An IPv6-only proxy fails when the target has no AAAA record. A /64-grouped rate limit looks like a block even when you rotate addresses inside the same range. Check which of the three you are seeing before changing your setup.
| Feature | IPv4 | IPv6 |
|---|---|---|
| Defining standard | RFC 791 (1981) | RFC 8200 (2017) |
| Address length | 32 bits | 128 bits |
| Number of addresses | About 4.3 billion | About 340 undecillion (3.4 × 10³⁸) |
| Notation | Four numbers from 0 to 255, separated by dots: 192.0.2.1 | Eight groups of hex digits separated by colons, with leading zeros dropped and one run of zero groups shortened to :: |
| Header size | 20 bytes minimum, up to 60 with options | Fixed 40 bytes, with optional extension headers for extra functions |
| Header checksum | Yes, recalculated at every hop | None; relies on link-layer and transport-layer checks |
| Who fragments packets | The sending host and any router on the path, unless the Don't Fragment flag is set | Only the sending host; routers drop oversized packets and send back an ICMPv6 Packet Too Big message |
| Minimum MTU | 68 bytes, with hosts required to accept 576-byte datagrams | 1,280 bytes |
| Delivery types | Unicast, broadcast, multicast (anycast only as a routing technique) | Unicast, multicast, and anycast |
| Address resolution | ARP (Address Resolution Protocol) | Neighbor Discovery Protocol (NDP), carried over ICMPv6 |
| Auto-configuration | Manual or DHCP, with a 169.254.0.0/16 link-local fallback | SLAAC builds an address from router advertisements; DHCPv6 is optional; every interface also gets an fe80:: link-local address |
| NAT | Widespread, including carrier-grade NAT, to stretch scarce addresses | Not needed, since every device can have a globally unique address |
| IPsec | Optional add-on | Originally mandatory to implement, recommended but optional since 2011 (RFC 6434, kept in RFC 8504) |
| DNS record type | A | AAAA |
| Loopback address | 127.0.0.1 (the whole 127.0.0.0/8 block is reserved) | ::1 |
| Private or local ranges | 10.0.0.0/8, 172.16.0.0/12 and 192.168.0.0/16; 169.254.0.0/16 link-local | fc00::/7 unique local, with fd00::/8 used in practice; fe80::/10 link-local |
Technical reference table; values are standards-based, not a live measurement.
| Myth / Claim | Facts (What's actually true) |
|---|---|
| IPsec is mandatory in IPv6. | Early IPv6 node requirements made IPsec mandatory. RFC 6434 relaxed that to should in 2011, and RFC 8504 kept it. Using IPsec was never required, and it works over IPv4 too. |
| NAT keeps IPv4 safe. | NAT blocks unsolicited incoming traffic only as a side effect of how it tracks connections. The IETF states that NAT was not developed as a security mechanism. A stateful firewall gives the same protection on IPv6, and neither stops malware a user downloads or a phishing link they click. |
| IPv6 can't be scanned. | Brute-forcing a whole /64 subnet is infeasible, but attackers narrow the search with predictable addresses such as 2001:db8::1 and 2001:db8::2, DNS records, reverse-DNS lookups, and addresses that embed an IPv4 address or a port number (RFC 7707). A server with a public hostname is as easy to find as it is on IPv4. |
| IPv6 addresses track you. | Early SLAAC built the device half of the address from the network card's hardware address, which followed a laptop from network to network. Modern systems use random identifiers and rotating temporary addresses (RFC 8981). The network prefix still identifies a household for as long as the ISP keeps it the same, much as a home's public IPv4 address does, and it can be more specific than an IPv4 address shared by many users through CGNAT. |
| You should disable IPv6 to be safe. | Microsoft advises against it: We don't recommend that you disable IPv6 or IPv6 components or unbind IPv6 from interfaces. If you do, some Windows components might not function. Its recommended alternative is to prefer IPv4 over IPv6 through prefix policies (the DisabledComponents registry value 0x20). The safer fix is to apply the same firewall rules to IPv6 as to IPv4. |
| Your VPN or proxy covers IPv6 too. | A VPN that only tunnels IPv4, or a system-wide proxy set up the same way, leaves IPv6 connections to dual-stack sites going out directly, in the clear, revealing your real IPv6 address (RFC 7359). Check that your provider either tunnels IPv6 or blocks it, and confirm with an IPv6 leak test. |
Security claims table; reflects standards and vendor guidance as cited, not a live audit.
| Measurement | Value | Source | Date |
|---|---|---|---|
| Users reaching Google over native IPv6 | 50.10%, first time above 50% | Google via Internet Society Pulse | March 28, 2026 |
| Worldwide IPv6 capability | 42% | APNIC | April 2026 |
| Share of requests to dual-stack sites over IPv6 | Less than a third in 2025, and over two-thirds in India | Cloudflare Radar Year in Review | 2025 |
| Websites using IPv6 | 31.8% | W3Techs | October 2026 |
| Top 1 million web hosts | 35.7% publish an IPv6 address and 31.7% answered over IPv6 from all five test locations | Common Crawl | July 2026 |
| Country examples | France 73% and India 72% on Google's measure | Internet Society Pulse | April 2026 |
Dated reference figures from different measurement methods; check current sources before relying on them.
| Site | IPv6 address | Loaded over IPv6 | Same response as IPv4 |
|---|---|---|---|
| Yes | Yes | Yes | |
| YouTube | Yes | Yes | Yes |
| Wikipedia | Yes | Yes | Yes |
| Amazon | Yes | Yes | Yes |
| eBay | No | No | n/a |
Selected targets from the October 2026 test; availability can change.
Frequently asked questions
Should I use IPv4 or IPv6?
Use both. Most devices and networks run dual stack, so traffic goes over IPv6 when the destination supports it and falls back to IPv4 when it does not. There is no need to pick one unless something breaks, and even then fixing the IPv6 problem beats switching IPv6 off.
Is IPv6 better for gaming?
Sometimes. IPv6 can sidestep carrier-grade NAT, which frequently produces strict or moderate NAT types that hinder hosting and voice chat. That benefit only holds for games and consoles that support IPv6, though, and many matchmaking services still run on IPv4, so results vary by title and network.
Should I enable IPv6 on my router?
Yes, provided your ISP offers IPv6. Current routers ship with a firewall that rejects unsolicited inbound IPv6 connections unless you change it. With IPv6 switched on, sites and apps that support it can use the newer protocol, and dual stack keeps IPv4 available for everything else, so nothing that works today stops working.
Should I disable IPv6?
In most cases, no. Microsoft's guidance is to make Windows prefer IPv4 over IPv6 rather than switch IPv6 off, since certain Windows features depend on it. When something misbehaves, set IPv4 as the preferred protocol or inspect the router and ISP settings before disabling anything. Turning IPv6 off can break features and merely conceals the real cause.
What are the disadvantages of IPv6?
Most of them are practical rather than technical. Coverage is uneven: some websites, game services, and corporate networks still do not support IPv6. Its addresses are longer and harder to read or type, and operating two protocols means maintaining two sets of firewall rules and monitoring. A poorly configured IPv6 setup can also introduce slow fallbacks or VPN leaks.
Why do some people still prefer IPv4?
Familiarity and compatibility. IPv4 addresses are short and easy to read; every network and tool supports them, and many administrators know their quirks well. Some older equipment, apps, and services still work only over IPv4, so it remains the safe default for anything that has to reach the whole internet.
Can IPv4 and IPv6 talk to each other directly?
No. They are separate protocols with different packet formats, so an IPv4-only device cannot talk to an IPv6-only one without help. Networks bridge them with dual stack, which runs both at once, or with translation such as NAT64 and DNS64, which lets IPv6-only devices reach IPv4-only sites.
How do I know if I have IPv6?
Visit a test site such as test-ipv6.com, or run curl -6 https://ifconfig.co in a terminal. If you get an IPv6 address back, your connection supports it. To see the addresses on your own machine, run ip -6 addr on Linux, ifconfig on macOS, or ipconfig on Windows.
Are IPv6 proxies good for web scraping?
They can be, for the right targets. IPv6 proxies are cheap and plentiful, but they only work on sites that support IPv6, and large sites often treat a whole /64 range as one client. They suit lightly protected, IPv6-ready targets; for hardened sites, IPv4 datacenter or residential proxies are more reliable.
Why are IPv6 proxies so cheap?
Because IPv6 addresses cost almost nothing to get. A single /48 allocation contains 65,536 /64 subnets, while IPv4 addresses sell for around $20 each and cloud providers charge for every public IPv4 address. The catch is that many addresses from one range add up to far fewer distinct identities than the count suggests.
curl -4 https://ifconfig.co
curl -6 https://ifconfig.co
curl.exe -4 https://ifconfig.co
curl.exe -6 https://ifconfig.co
curl -6 -sS -o /dev/null -w "%{http_code} %{remote_ip}\n" https://example.com
| Sample | Websites tested | Publish an IPv6 address (AAAA record) | Loaded over IPv6 from both servers |
|---|---|---|---|
| Tranco top 1,000 | 725 | 54.1% | 49.1% |
| Commonly scraped sites | 44 | 31.8% | 29.5% |
Test run October 9-10, 2026 from two servers; figures are a dated snapshot, not a live measurement.