Residential Proxy Benchmark 2026: 12 Providers Tested

We bought a residential package from each of 12 providers and ran every test through the same scripts, servers and schedule: success rate, latency, duplicate IPs, sticky sessions, Amazon access, IP quality and cost per successful request.

We operate Floppydata and tested it on the same terms as everyone else, including where it came out behind.

See the results

12 providers

12 exit countries

302,400 requests

5–8 Oct 2026

Residential Benchmark

Results at a glance

Residential proxy benchmark results for 12 providers: Floppydata first, then by Success Rate
Provider	Cost per 1,000 successful requests ($)	Price ($/GB)	Proxy Detection Rate	Fraud Score avg / high-risk	Latency median / P90 (s)	Amazon Target Success Rate	Success Rate	Duplicate Rate	Sticky Session Stability
Floppydata	0.019	1.00	60.7%	3.0 / 0.3%	0.93 / 1.76	49.4%	98.0%	11.7%	47.3%
Decodo	0.077	3.75	85.0%	3.0 / 0.3%	0.79 / 1.41	54.4%	99.9%	2.2%	82.7%
Oxylabs	0.124	6.00	80.3%	4.6 / 1.3%	0.75 / 1.32	55.3%	99.9%	2.2%	80.8%
Evomi	0.020	1.04	96.0%	3.3 / 0.3%	0.68 / 1.44	49.4%	99.9%	5.2%	80.4%
SOAX	0.091	4.67	85.3%	3.3 / 0.3%	0.72 / 1.37	49.5%	99.8%	9.1%	70.2%
DataImpulse	0.020	1.00	87.0%	3.3 / 0.7%	0.87 / 2.08	47.8%	99.7%	5.3%	87.2%
Rayobyte	0.066	3.50	59.3%	4.4 / 1.3%	1.45 / 3.00	51.1%	99.7%	22.5%	70.2%
ProxyEmpire	0.166	8.47	83.3%	3.7 / 0.3%	0.83 / 1.83	49.7%	99.6%	5.4%	61.2%
Webshare	0.052	2.70	83.3%	3.6 / 0.0%	1.00 / 1.81	54.6%	99.6%	2.2%	77.2%
NodeMaven	0.100	5.00	77.7%	2.3 / 0.0%	0.91 / 1.71	50.2%	99.6%	6.7%	79.2%
IPRoyal	0.104	5.31	96.0%	4.5 / 1.0%	1.23 / 3.14	50.1%	99.5%	7.0%	80.4%
Geonode	0.015	0.79	99.3%	3.7 / 0.3%	1.09 / 3.67	49.3%	97.1%	6.6%	80.4%

Top 3 in the column Floppydata (we operate it; listed first, other providers by Success Rate)

Amazon Target Success Rate swung between roughly 28% and 83% from run to run for every provider, which points to Amazon-side behaviour rather than provider quality. Use it to compare providers within this test, not as an absolute figure. A lower Proxy Detection Rate is better: fewer IPs were flagged as proxies.

Check the numbers yourself: the anonymized raw data, scripts and full results workbook are on GitHub.

Our methodology

How we tested

We bought a residential package from each of 12 providers at the price listed on its website and ran every test through the same scripts. We tested between 5 and 8 October 2026. Bright Data and Byteful were not included because they require identity verification (KYC).

Servers. Three Hetzner cloud servers in Germany, the US and Singapore. Each ran the same scripts on a fixed schedule (cron, UTC).
Countries. 12 exit countries: US, DE, GB, FR, ES, BR, IN, PL, UA, TR, VN, ID.
Providers. Decodo, Oxylabs, Evomi, SOAX, DataImpulse, Rayobyte, ProxyEmpire, Webshare, NodeMaven, IPRoyal, Floppydata, Geonode.

Tests

Infrastructure test (Success Rate, Latency, Duplicate Rate). Each server sent 100 requests per provider and country through rotating residential proxies to a Cloudflare trace endpoint, at 08:00 and 20:00 UTC every day. Success Rate is the share of requests that returned a valid answer. Latency is the median and 90th percentile (P90) of response time on successful requests. Duplicate Rate is 1 minus unique IPs divided by successful requests. Confidence intervals use the Wilson 95% method.
Target test (Target Success Rate). Amazon (US, DE, GB) and Wikipedia were requested at 11:00 and 23:00 UTC with a plain HTTP client that uses a Chrome TLS fingerprint and no JavaScript. A request counts as a success only when the real page came back, not a CAPTCHA or block page.
Sticky session test (Sticky Session Stability). Each run opened 8 sticky sessions per provider and checked them every 30 seconds for 30 minutes. A session counts as stable if the IP never changed and at least 90% of checks were answered during the first 25 minutes.
Time to first byte. On 7 October we measured TTFB with curl: 10 requests per provider, country and server.
IP quality. We checked a sample of 300 unique IPs per provider (3,600 in total) with the Scamalytics API (fraud score, proxy detection) and an ASN check with IPinfo (residential or ISP = not flagged as datacenter and not on a hosting network).
Cost per Successful Request. We recorded each provider's price per GB and read the traffic counter in each provider's dashboard on 8 October at 14:50 UTC. Cost per Successful Request = price per GB × GB used ÷ successful requests of all our tests up to that moment. We report it per 1,000 requests.

Limitations and disclosure

Disclosure. We operate Floppydata, one of the 12 providers in this test. We ran it through the same scripts, servers and schedule as the other 11 providers, and report its results, including the weaker ones, in the same way. Our Floppydata account was topped up by the company, but the price used in the cost calculation is the public price of $1.00/GB, the same for all users. Floppydata is the lowest in sticky stability (47.3%) and second-lowest in Success Rate (98.0%), and we say so in the key findings.

Short period. The test covers four days (5–8 October 2026). Provider quality can change with pool size, time of day and load.
Three test locations. All requests came from servers in Germany, the US and Singapore. Results from other locations or networks may differ.
Plain HTTP client for targets. The Amazon and Wikipedia test used no JavaScript and no browser. Amazon results moved between about 28% and 83% from run to run for every provider, so they should be read as a comparison inside this test, not as an absolute success rate.
One source for IP quality. Fraud Score and Proxy Detection Rate come from Scamalytics, and network names (ASN) and geolocation from IPinfo, on a sample of 300 IPs per provider. Another database would rate the same IPs differently.
Prices as paid. Prices are what we paid for each package, which differ in size (from 2 GB to 50 GB), tariff and tax. ProxyEmpire's $8.47/GB includes tax ($7.00 without). Evomi's price is converted from EUR.
Cost measured by counters. Cost per Successful Request uses each provider's own traffic counter, read on 8 October at 14:50 UTC. Counters may count traffic differently (1 GB can be 1,000 or 1,024 MB), which moves the result by a few percent.
Sticky sessions need each provider's own format. We used the session format from each provider's documentation. A different parameter setup may give a different result. For Floppydata the format was corrected on 6 October, and earlier sessions are excluded.
Not tested. Bright Data and Byteful were left out because they require identity verification. Mobile and datacenter proxies are not covered.
Not a recommendation. The data compares providers on a specific set of tests. Choose by the task you run.

Raw data and reproducibility

The test used five Python scripts, one for each test: infrastructure (bench.py), targets (targets.py), sticky sessions (sticky.py), TTFB (ttfb.py) and IP quality (ipquality.py), plus analyze.py, which builds the results workbook from the raw CSV logs. The workbook contains the full per-server, per-country and per-run tables, confidence intervals and the method notes.

Because Floppydata is one of the tested providers, we publish the raw data, the scripts and the results workbook on GitHub, so anyone can check the numbers. The raw logs are anonymized: every exit IP address is replaced by a keyed hash. The same address always gets the same hash, so unique-IP counts, Duplicate Rate and sticky-session IP changes can be recomputed, but the addresses cannot be recovered. No credentials, API keys or server addresses are published.

Repository. Floppydata/residential-proxy-benchmark-2026: data, scripts and the method in one place.
README. What was tested, how the data was anonymized, column reference and how to recompute the results.
Raw request logs. 87 CSV files, one row per request or check (about 636,000 rows), split by test server: us, de, sg.
IP quality results. Per-provider aggregates only. Individual IPs and raw Scamalytics responses are not published.
Scripts. The five test scripts and analyze.py. Running analyze.py on the published CSV files gives the same tables as the published workbook, except IP quality.
Results workbook. results.xlsx with all tables, confidence intervals, per-run trends and method notes.

Code is under the MIT licence, data and workbook under CC BY 4.0. To run your own test you need your own proxy credentials and, for IP quality, your own Scamalytics and IPinfo access.

Full results

Full residential proxy benchmark results with 95% confidence intervals and per-server values: Floppydata first
Provider	Cost per 1,000 successful requests ($)	Price ($/GB)	Proxy Detection Rate	IP Risk / Fraud Score	Residential / ISP Classification	Geo Accuracy	Latency median / P90 (s)	Latency median US / DE / SG (s)	TTFB median US / DE / SG (s)	Unique IPs	Duplicate Rate	Success Rate	Success Rate US / DE / SG	Weakest country	Error Rate (CAPTCHA / Block Rate)	Target Success Rate	Sticky Session Stability
Floppydata	0.019 est. split: Infra 0.0063 · Wikipedia 0.067 · Amazon 0.375	1.00 50 GB, 30 days	60.7% (55.0–66.0)	avg 3.0 · median 1.0 · high risk 0.3% (0.1–1.9)	99.0% (97.1–99.7)	99.8% (99.7–99.8)	0.93 / 1.76	1.21 / 0.78 / 0.73	1.36 / 0.81 / 0.77	21,813 of 24,703	11.7%	98.0% (97.8–98.2)	97.8% / 97.7% / 98.6%	IN 90.0%	Infra 2.0% · Amazon: CAPTCHA 31.3%, JS challenge 17.7%, blocked 0.0%	Wikipedia 98.6% (98.2–98.8) · Amazon 49.4% (47.2–51.7)	47.3% (41.4–53.4), n=264
Decodo	0.077 est. split: Infra 0.0233 · Wikipedia 0.250 · Amazon 1.368	3.75 3 GB	85.0% (80.5–88.6)	avg 3.0 · median 1.0 · high risk 0.3% (0.1–1.9)	99.3% (97.6–99.8)	99.8% (99.8–99.9)	0.79 / 1.41	0.94 / 0.54 / 0.80	0.92 / 0.44 / 0.82	24,618 of 25,184	2.2%	99.9% (99.9–100.0)	100.0% / 99.9% / 99.9%	IN 99.8%	Infra 0.1% · Amazon: CAPTCHA 35.3%, JS challenge 10.2%, blocked 0.0%	Wikipedia 99.9% (99.7–99.9) · Amazon 54.4% (52.1–56.6)	82.7% (78.1–86.5), n=312
Oxylabs	0.124 est. split: Infra 0.0373 · Wikipedia 0.400 · Amazon 2.184	6.00 5 GB	80.3% (75.5–84.4)	avg 4.6 · median 1.0 · high risk 1.3% (0.5–3.4)	99.3% (97.6–99.8)	99.9% (99.8–99.9)	0.75 / 1.32	0.76 / 0.56 / 0.80	0.77 / 0.56 / 0.85	24,628 of 25,180	2.2%	99.9% (99.9–99.9)	99.9% / 100.0% / 99.9%	IN 99.7%	Infra 0.1% · Amazon: CAPTCHA 34.2%, JS challenge 10.3%, blocked 0.0%	Wikipedia 99.8% (99.6–99.9) · Amazon 55.3% (53.1–57.6)	80.8% (76.0–84.8), n=312
Evomi	0.020 est. split: Infra 0.0065 · Wikipedia 0.069 · Amazon 0.384	1.04 Core tariff, EUR 0.93 ex-VAT	96.0% (93.1–97.7)	avg 3.3 · median 1.0 · high risk 0.3% (0.1–1.9)	99.3% (97.6–99.8)	99.6% (99.5–99.7)	0.68 / 1.44	0.41 / 0.38 / 0.97	0.40 / 0.33 / 0.93	23,867 of 25,163	5.2%	99.9% (99.8–99.9)	99.8% / 99.8% / 99.9%	VN 99.6%	Infra 0.1% · Amazon: CAPTCHA 32.2%, JS challenge 18.3%, blocked 0.0%	Wikipedia 99.7% (99.6–99.8) · Amazon 49.4% (47.1–51.6)	80.4% (75.7–84.5), n=312
SOAX	0.091 est. split: Infra 0.0291 · Wikipedia 0.312 · Amazon 1.743	4.67 effective price, 1 credit = $1	85.3% (80.9–88.9)	avg 3.3 · median 1.0 · high risk 0.3% (0.1–1.9)	99.0% (97.1–99.7)	99.3% (99.2–99.4)	0.72 / 1.37	0.64 / 0.56 / 0.80	0.79 / 0.50 / 0.85	22,866 of 25,152	9.1%	99.8% (99.7–99.9)	99.8% / 99.8% / 99.8%	UA 99.5%	Infra 0.2% · Amazon: CAPTCHA 31.6%, JS challenge 17.5%, blocked 0.0%	Wikipedia 99.6% (99.4–99.8) · Amazon 49.5% (47.2–51.7)	70.2% (64.9–75.0), n=312
DataImpulse	0.020 est. split: Infra 0.0062 · Wikipedia 0.067 · Amazon 0.354	1.00 50 GB	87.0% (82.7–90.3)	avg 3.3 · median 1.0 · high risk 0.7% (0.2–2.4)	98.0% (95.7–99.1)	98.8% (98.7–98.9)	0.87 / 2.08	0.72 / 0.75 / 0.95	0.82 / 0.59 / 1.04	23,795 of 25,125	5.3%	99.7% (99.6–99.8)	99.7% / 99.7% / 99.7%	TR 98.9%	Infra 0.3% · Amazon: CAPTCHA 31.6%, JS challenge 18.6%, blocked 0.0%	Wikipedia 99.6% (99.4–99.8) · Amazon 47.8% (45.6–50.1)	87.2% (83.0–90.4), n=312
Rayobyte	0.066 est. split: Infra 0.0218 · Wikipedia 0.234 · Amazon 1.285	3.50 2 GB	59.3% (53.7–64.7)	avg 4.4 · median 1.0 · high risk 1.3% (0.5–3.4)	95.0% (91.9–96.9)	98.9% (98.7–99.0)	1.45 / 3.00	0.81 / 1.29 / 1.80	0.96 / 1.21 / 1.85	19,470 of 25,126	22.5%	99.7% (99.6–99.8)	99.9% / 99.7% / 99.6%	UA 99.1%	Infra 0.3% · Amazon: CAPTCHA 32.4%, JS challenge 16.1%, blocked 0.0%	Wikipedia 99.6% (99.4–99.7) · Amazon 51.1% (48.9–53.4)	70.2% (64.9–75.0), n=312
ProxyEmpire	0.166 est. split: Infra 0.0529 · Wikipedia 0.565 · Amazon 3.158	8.47 2 x 1 GB, $7.00 ex-tax, 8.47 incl. tax	83.3% (78.7–87.1)	avg 3.7 · median 1.0 · high risk 0.3% (0.1–1.9)	99.3% (97.6–99.8)	99.7% (99.6–99.8)	0.83 / 1.83	0.73 / 0.64 / 1.18	0.78 / 0.57 / 1.40	23,738 of 25,092	5.4%	99.6% (99.5–99.6)	99.5% / 99.5% / 99.7%	TR 97.7%	Infra 0.4% · Amazon: CAPTCHA 32.3%, JS challenge 17.4%, blocked 0.0%	Wikipedia 99.4% (99.2–99.6) · Amazon 49.7% (47.4–51.9)	61.2% (55.7–66.5), n=312
Webshare	0.052 est. split: Infra 0.0168 · Wikipedia 0.180 · Amazon 0.986	2.70 3 GB	83.3% (78.7–87.1)	avg 3.6 · median 1.0 · high risk 0.0% (0.0–1.3)	99.3% (97.6–99.8)	99.8% (99.8–99.9)	1.00 / 1.81	0.90 / 0.71 / 1.33	1.04 / 0.66 / 1.29	24,545 of 25,091	2.2%	99.6% (99.5–99.6)	99.9% / 99.7% / 99.2%	PL 99.0%	Infra 0.4% · Amazon: CAPTCHA 33.3%, JS challenge 11.5%, blocked 0.0%	Wikipedia 99.4% (99.2–99.6) · Amazon 54.6% (52.4–56.8)	77.2% (72.3–81.5), n=312
NodeMaven	0.100 est. split: Infra 0.0312 · Wikipedia 0.334 · Amazon 1.858	5.00 2 GB	77.7% (72.6–82.0)	avg 2.3 · median 1.0 · high risk 0.0% (0.0–1.3)	99.7% (98.1–99.9)	99.8% (99.7–99.8)	0.91 / 1.71	1.00 / 0.64 / 1.13	0.99 / 0.54 / 0.97	23,411 of 25,098	6.7%	99.6% (99.5–99.7)	99.6% / 99.5% / 99.7%	US 98.9%	Infra 0.4% · Amazon: CAPTCHA 32.6%, JS challenge 16.6%, blocked 0.0%	Wikipedia 99.2% (98.9–99.4) · Amazon 50.2% (47.9–52.4)	79.2% (74.3–83.3), n=312
IPRoyal	0.104 est. split: Infra 0.0332 · Wikipedia 0.354 · Amazon 1.921	5.31 2 GB, may include tax	96.0% (93.1–97.7)	avg 4.5 · median 1.0 · high risk 1.0% (0.3–2.9)	99.3% (97.6–99.8)	99.6% (99.5–99.6)	1.23 / 3.14	1.21 / 0.92 / 1.38	1.05 / 1.00 / 1.62	23,316 of 25,063	7.0%	99.5% (99.4–99.5)	99.1% / 99.6% / 99.6%	US 98.9%	Infra 0.5% · Amazon: CAPTCHA 32.2%, JS challenge 16.8%, blocked 0.0%	Wikipedia 99.6% (99.3–99.7) · Amazon 50.1% (47.9–52.4)	80.4% (75.7–84.5), n=312
Geonode	0.015 est. split: Infra 0.0051 · Wikipedia 0.053 · Amazon 0.287	0.79 10 GB, subscription	99.3% (97.6–99.8)	avg 3.7 · median 1.0 · high risk 0.3% (0.1–1.9)	100.0% (98.7–100.0)	99.3% (99.2–99.4)	1.09 / 3.67	1.09 / 0.81 / 1.37	1.72 / 2.22 / 1.90	22,836 of 24,459	6.6%	97.1% (96.8–97.3)	96.9% / 97.4% / 96.9%	IN 94.6%	Infra 2.9% · Amazon: CAPTCHA 32.9%, JS challenge 15.3%, blocked 0.0%	Wikipedia 97.9% (97.5–98.3) · Amazon 49.3% (47.1–51.6)	80.4% (75.7–84.5), n=312

Pooled over three test servers (USA, Germany, Singapore). Figures in brackets are 95% Wilson confidence intervals. US / DE / SG are the values measured from each test server.

Weakest country is the requested exit country with the lowest Success Rate. The cost split by test is an estimate from modelled bytes per request, calibrated on the providers' traffic counters; the measured total covers all tests (1 GB = 1,000 MB).

Floppydata is listed first; the other providers follow in order of Success Rate.

Results by metric

Cost per 1,000

We did not model costs. We read each provider's traffic counter after the tests and divided the money spent on that traffic by the successful requests.

Because the test requests are small (Cloudflare trace, Wikipedia and Amazon pages), the cost follows the price per GB almost directly. A heavier workload costs more per request at every provider.

Cost per 1,000 successful requests ($) Lower is better

Geonode $0.015
Floppydata $0.019
Evomi $0.020
DataImpulse $0.020
Webshare $0.052
Rayobyte $0.066
Decodo $0.077
SOAX $0.091
NodeMaven $0.100
IPRoyal $0.104
Oxylabs $0.124
ProxyEmpire $0.166

Key findings

In our 12-provider test (5–8 October 2026), Geonode, Floppydata, DataImpulse and Evomi had the lowest cost per successful request, Rayobyte and Floppydata the fewest IPs flagged as proxies, and Floppydata the fastest connections from Singapore. Decodo, Oxylabs and Evomi had the highest Success Rate, and Evomi, SOAX and Oxylabs the lowest latency overall. We operate Floppydata and tested it on the same terms as everyone else; see the disclosure in the methodology.

Speed from Singapore

Floppydata had the lowest median latency (0.73 s) and the fastest median time to first byte (0.77 s) from our Singapore server, ahead of Oxylabs, SOAX and Decodo.
Cost per successful request

$0.015–$0.020 per 1,000 successful requests at Geonode, Floppydata, DataImpulse and Evomi, against $0.124 at Oxylabs and $0.166 at ProxyEmpire. Prices are as we paid for each package.
Proxy detection

Rayobyte (59.3%) and Floppydata (60.7%) had the fewest IPs flagged as proxies by Scamalytics; Geonode (99.3%) the most. Lower means fewer IPs flagged.
Price per GB

Geonode $0.79, DataImpulse and Floppydata $1.00, Evomi $1.04 — against $6.00 at Oxylabs and $8.47 at ProxyEmpire (with tax), as paid for each package.
Fraud score

Fraud scores are low everywhere (median 1.0 at all 12 providers). Lowest averages: NodeMaven 2.3, Decodo and Floppydata 3.0.
Geo accuracy

98.8–99.9% of exit IPs were in the requested country at every provider: Oxylabs 99.9%; Decodo, Floppydata, Webshare and NodeMaven 99.8%.
IP type

95–100% of the IPs we checked are residential or ISP addresses at every provider. Rayobyte had the most datacenter-flagged IPs (4.3%).
Latency overall

Median: Evomi 0.68 s, SOAX 0.72 s, Oxylabs 0.75 s. Slowest: Rayobyte 1.45 s and IPRoyal 1.23 s.
Success Rate

Decodo, Oxylabs and Evomi 99.9%; Floppydata 98.0% and Geonode 97.1% are the lowest. Floppydata's figure includes a one-off failure in India in the first run; without it, about 98.7%.
Sticky sessions

DataImpulse (87.2%), Decodo (82.7%) and Oxylabs (80.8%) held the same IP most often; Floppydata was lowest at 47.3%.
Duplicate IPs

Decodo, Oxylabs and Webshare repeated 2.2% of IPs among successful requests; Floppydata 11.7% and Rayobyte 22.5%.
Amazon

With a plain HTTP client and no JavaScript, 47.8–55.3% of Amazon requests returned a real page at every provider; Oxylabs, Webshare and Decodo were highest. The ranking is only indicative.

Provider by provider

Decodo $0.077 / 1k req · $3.75/GB · 0.79 s

Top Success Rate (99.9%), one of the lowest Duplicate Rates (2.2%), second-best sticky sessions (82.7%) and the third-highest Amazon result (54.4%). $3.75/GB.

Cost per 1,000 successful requests: $0.077
Price: $3.75/GB
Proxy Detection Rate: 85.0%
Fraud Score avg / high-risk: 3.0 / 0.3%
Latency median / P90: 0.79 / 1.41 s
Amazon Target Success Rate: 54.4%
Success Rate: 99.9%
Duplicate Rate: 2.2%
Sticky Session Stability: 82.7%

Oxylabs $0.124 / 1k req · $6.00/GB · 0.75 s

99.9% Success Rate, lowest P90 latency (1.32 s), highest Amazon result (55.3%). The most expensive in the group after ProxyEmpire at $6.00/GB (package of 5 GB).

Cost per 1,000 successful requests: $0.124
Price: $6.00/GB
Proxy Detection Rate: 80.3%
Fraud Score avg / high-risk: 4.6 / 1.3%
Latency median / P90: 0.75 / 1.32 s
Amazon Target Success Rate: 55.3%
Success Rate: 99.9%
Duplicate Rate: 2.2%
Sticky Session Stability: 80.8%

Evomi $0.020 / 1k req · $1.04/GB · 0.68 s

99.9% Success Rate and the lowest median latency (0.68 s) at $1.04/GB. Duplicate Rate 5.2%. A strong result for the price.

Cost per 1,000 successful requests: $0.020
Price: $1.04/GB
Proxy Detection Rate: 96.0%
Fraud Score avg / high-risk: 3.3 / 0.3%
Latency median / P90: 0.68 / 1.44 s
Amazon Target Success Rate: 49.4%
Success Rate: 99.9%
Duplicate Rate: 5.2%
Sticky Session Stability: 80.4%

SOAX $0.091 / 1k req · $4.67/GB · 0.72 s

99.8% Success Rate and fast (0.72 s median), but a Duplicate Rate of 9.1% and sticky stability of 70.2%. Effective price $4.67/GB.

Cost per 1,000 successful requests: $0.091
Price: $4.67/GB
Proxy Detection Rate: 85.3%
Fraud Score avg / high-risk: 3.3 / 0.3%
Latency median / P90: 0.72 / 1.37 s
Amazon Target Success Rate: 49.5%
Success Rate: 99.8%
Duplicate Rate: 9.1%
Sticky Session Stability: 70.2%

DataImpulse $0.020 / 1k req · $1.00/GB · 0.87 s

Best sticky stability (87.2%) and 99.7% Success Rate at $1.00/GB. Slower at P90 (2.08 s).

Cost per 1,000 successful requests: $0.020
Price: $1.00/GB
Proxy Detection Rate: 87.0%
Fraud Score avg / high-risk: 3.3 / 0.7%
Latency median / P90: 0.87 / 2.08 s
Amazon Target Success Rate: 47.8%
Success Rate: 99.7%
Duplicate Rate: 5.3%
Sticky Session Stability: 87.2%

Rayobyte $0.066 / 1k req · $3.50/GB · 1.45 s

99.7% Success Rate, but the slowest median latency (1.45 s) and the highest Duplicate Rate (22.5%). Lowest Proxy Detection Rate (59.3%) and the largest share of datacenter-flagged IPs (4.3%).

Cost per 1,000 successful requests: $0.066
Price: $3.50/GB
Proxy Detection Rate: 59.3%
Fraud Score avg / high-risk: 4.4 / 1.3%
Latency median / P90: 1.45 / 3.00 s
Amazon Target Success Rate: 51.1%
Success Rate: 99.7%
Duplicate Rate: 22.5%
Sticky Session Stability: 70.2%

ProxyEmpire $0.166 / 1k req · $8.47/GB · 0.83 s

99.6% Success Rate and 0.83 s median latency, with the lowest sticky stability among the established providers (61.2%). The highest price ($8.47/GB with tax).

Cost per 1,000 successful requests: $0.166
Price: $8.47/GB
Proxy Detection Rate: 83.3%
Fraud Score avg / high-risk: 3.7 / 0.3%
Latency median / P90: 0.83 / 1.83 s
Amazon Target Success Rate: 49.7%
Success Rate: 99.6%
Duplicate Rate: 5.4%
Sticky Session Stability: 61.2%

Webshare $0.052 / 1k req · $2.70/GB · 1.00 s

99.6% Success Rate, 2.2% Duplicate Rate and no high-risk IPs in the sample. Median latency 1.00 s. $2.70/GB.

Cost per 1,000 successful requests: $0.052
Price: $2.70/GB
Proxy Detection Rate: 83.3%
Fraud Score avg / high-risk: 3.6 / 0.0%
Latency median / P90: 1.00 / 1.81 s
Amazon Target Success Rate: 54.6%
Success Rate: 99.6%
Duplicate Rate: 2.2%
Sticky Session Stability: 77.2%

NodeMaven $0.100 / 1k req · $5.00/GB · 0.91 s

99.6% Success Rate, the lowest average fraud score (2.3) and no high-risk IPs. $5.00/GB.

Cost per 1,000 successful requests: $0.100
Price: $5.00/GB
Proxy Detection Rate: 77.7%
Fraud Score avg / high-risk: 2.3 / 0.0%
Latency median / P90: 0.91 / 1.71 s
Amazon Target Success Rate: 50.2%
Success Rate: 99.6%
Duplicate Rate: 6.7%
Sticky Session Stability: 79.2%

IPRoyal $0.104 / 1k req · $5.31/GB · 1.23 s

99.5% Success Rate with a P90 latency of 3.14 s. Sticky stability 80.4%. $5.31/GB.

Cost per 1,000 successful requests: $0.104
Price: $5.31/GB
Proxy Detection Rate: 96.0%
Fraud Score avg / high-risk: 4.5 / 1.0%
Latency median / P90: 1.23 / 3.14 s
Amazon Target Success Rate: 50.1%
Success Rate: 99.5%
Duplicate Rate: 7.0%
Sticky Session Stability: 80.4%

Floppydata $0.019 / 1k req · $1.00/GB · 0.93 s

98.0% Success Rate, 0.93 s median latency, a Duplicate Rate of 11.7%, and the lowest sticky stability in the test (47.3%). Cost per Successful Request is among the three lowest ($0.019 per 1,000). We operate this provider; see the disclosure above.

Cost per 1,000 successful requests: $0.019
Price: $1.00/GB
Proxy Detection Rate: 60.7%
Fraud Score avg / high-risk: 3.0 / 0.3%
Latency median / P90: 0.93 / 1.76 s
Amazon Target Success Rate: 49.4%
Success Rate: 98.0%
Duplicate Rate: 11.7%
Sticky Session Stability: 47.3%

Geonode $0.015 / 1k req · $0.79/GB · 1.09 s

The lowest price ($0.79/GB) and the lowest cost per 1,000 requests ($0.015), with a 97.1% Success Rate, the highest P90 latency (3.67 s) and the slowest median TTFB (1.98 s).

Cost per 1,000 successful requests: $0.015
Price: $0.79/GB
Proxy Detection Rate: 99.3%
Fraud Score avg / high-risk: 3.7 / 0.3%
Latency median / P90: 1.09 / 3.67 s
Amazon Target Success Rate: 49.3%
Success Rate: 97.1%
Duplicate Rate: 6.6%
Sticky Session Stability: 80.4%

Frequently Asked Questions

Which provider was fastest from Singapore?

Floppydata. From our Singapore server it had the lowest median latency (0.73 s, ahead of Oxylabs, SOAX and Decodo at 0.80 s) and the fastest median time to first byte (0.77 s, then Decodo at 0.82 s). From the US server it was slower (1.21 s median latency).

Which provider has the lowest cost per successful request?

Geonode ($0.015 per 1,000 successful requests), Floppydata ($0.019), DataImpulse ($0.020) and Evomi ($0.020). This is measured from provider traffic counters on small test requests and follows the price per GB.

Which provider's IPs were flagged as proxies least often?

Rayobyte (59.3%) and Floppydata (60.7%) had the fewest IPs flagged as proxies by Scamalytics, and Geonode (99.3%) the most. In our sample, 95% to 100% of IPs were residential or ISP addresses, and fraud scores were low at every provider.

Which provider was fastest overall?

Evomi had the lowest median latency (0.68 s), followed by SOAX (0.72 s) and Oxylabs (0.75 s). Oxylabs had the lowest P90 (1.32 s).

Which residential proxy provider had the highest Success Rate?

Decodo, Oxylabs and Evomi, each at 99.9% in the infrastructure test (25,200 requests each). Seven more providers finished between 99.5% and 99.8%.

Which provider had the most stable sticky sessions?

DataImpulse (87.2%), then Decodo (82.7%) and Oxylabs (80.8%).

Why is the Amazon success rate so low?

Our test used a plain HTTP client without JavaScript, and Amazon answered about a third of requests with a CAPTCHA and another 10–19% with a JavaScript challenge. The rate also swung between runs for all providers, so use it for comparison only.

Did you test your own provider fairly?

We operate Floppydata and tested it with the same scripts, servers, countries and schedule as every other provider. It ranked lowest on sticky sessions and second-lowest on Success Rate. See the disclosure above.

Residential Proxies

Datacenter Proxies

Mobile Proxies

Static Residential Proxies

Static Datacenter Proxies

Find Your Proxy

Popular Locations

World Location

Scraping API

Popular Use Cases

For Social Media

Cloud Browser

Cloud Browser

More Solutions

Popular Use Cases

For Social Media

For Use Cases

Free Tools

Resources

Join our community

Language

Residential Proxies

Datacenter Proxies

Mobile Proxies

Static Residential

Static Datacenter

Find Your Proxy

Popular Locations

World Location

Scraping API

Cloud Browser

Popular Use Cases

Free Tools

Resources

Residential Proxy Benchmark 2026: 12 Providers Tested

Results at a glance

Our methodology

How we tested

Tests

Limitations and disclosure

Raw data and reproducibility

Download raw benchmark data

Full results

Results by metric

Cost per 1,000

Price

Proxy Detection

Fraud Score

Residential / ISP

Geo Accuracy

Latency

Latency by server

TTFB by server

Amazon

Wikipedia

Success Rate

Success by server

Weakest country

Error Rate

Duplicate Rate

Sticky Sessions

Key findings

Speed from Singapore

Cost per successful request

Proxy detection

Price per GB

Fraud score

Geo accuracy

IP type

Latency overall

Success Rate

Sticky sessions

Duplicate IPs

Amazon

Provider by provider

Frequently Asked Questions

Run your own test with Floppydata residential proxies