Free China proxy list 2026

The table compares live Chinese entries with every other live entry, measured in the same minute. An address carrying 20 or more live entries counts as busy, and the table leaves it out.

The success lines matter most: live Chinese entries pass far more of their own checks than the rest. With its two busy addresses counted, the Chinese median falls to 62.9 percent. The latency line is measured from the United States and includes the distance to China.

We tested every row of the public China list twice on 28 September, at 15:11 and 15:45 UTC. Each entry was asked for four things, in its own protocol, from our server in the United States. An echo returned the exit address, a Chinese lookup service placed it, and we opened Baidu and Google.

None of the rows that exited in mainland China opened Google, in either run. Four rows opened it each time; they exited abroad or never showed an exit.

The Chinese lookup service agreed with the echo on 82 of the 84 addresses it answered. One row showed the echo an American exit while the lookup service saw a Chinese one.

Across both runs, 19 of the 116 relays left through an address other than the listed one. Four rows presented certificates that did not verify; one did so for both sites in both runs. That is what interception of encrypted traffic looks like.

Two addresses carried 449 of the rows, one from Alibaba and one from Abitcool. The Alibaba address relayed 7 rows in the first run and 112 in the second. Nearly every relay left through a different Chinese address, the mark of a gateway. The Abitcool address relayed 3 rows, then none.

Most live Chinese entries sit on the networks of internet providers, not in data centers. The operator names come from the APNIC registry record for each network number.

Alibaba is the exception: all 85 of its entries sit on datacenter space, which means rented cloud servers. 53 of them shared one address, each on its own port.

Most Chinese entries never answer at all: 17,892 of the 20,628 set aside from ordinary addresses, 86.7 percent.

The rest of the pool fares slightly worse, at 88.7 percent. The gap opens among the entries that do work, timed below from first sighting to last answer.

Half the counted Chinese entries lasted 26 days or more, about three times the rest's 8.8.

Without the 118 on eight addresses that hold 20 or more entries overall, the median is 23.7 days.

If you are in mainland China, this list will not get you past the firewall. A Chinese entry exits inside China, so its traffic meets the same filters as yours.

Those filters act on domain names as well as on the traffic itself. A study over nine months counted 311,000 domains censored by the firewall's DNS filter (USENIX Security, 2021). Since November 2021 it has also detected and blocked fully encrypted proxy traffic (USENIX Security, 2023).

The rules matter too: a 2017 notice from China's Ministry of Industry and Information Technology covers cross-border channels. It prohibits setting up or leasing dedicated lines, VPNs included, for cross-border business without approval. This page only reports what the notice says, and it is not legal advice.

Copy a row's address and port, and set the matching protocol in your client. Browsers and scrapers can use HTTP or HTTPS entries. SOCKS5 entries also work with programs that do not speak HTTP.

Test an entry against a Chinese site before real work, because from abroad these proxies are slow. From the United States, our checker measured a median of 4,845 ms; none answered within 500 ms.

Take several entries and rotate between them, rather than trusting any single address to stay up. A live Chinese entry passed a median 78.9 percent of its checks, so expect failures.

A free SOCKS5 entry adds no encryption of its own to what you send. Under the SOCKS5 standard, RFC 1928, encryption depends on the method both sides agree on. Free entries use the method with no authentication, which carries no encryption either. Never send a password or a payment detail through one.