Proxy marketing claims are worthless. "Millions of IPs", "99% uptime", "best-in-class performance" - every provider says the same thing. We decided to actually measure it. Over four weeks, we ran 50,000 requests through the top residential proxy providers and published the real numbers.
Test Methodology
We wanted numbers that reflect real-world usage, not cherry-picked benchmarks. Here's exactly what we did:
- Total requests: 50,000 per provider (200,000+ total)
- Target sites: 5 major sites with bot protection - an e-commerce platform, a travel aggregator, a social media site, a e-commerce retailer, and a financial data site
- Geo targets: US, UK, Germany, Japan - 25% each
- Test tool: Python
requestslibrary + custom Scrapy spider - Success criteria: HTTP 200 with expected content (not a CAPTCHA or block page)
- Measurement period: 4 weeks, with requests distributed across all hours
import requestsimport timeimport statisticsfrom datetime import datetime
def test_proxy(proxy_url, target, n=1000): """Test a proxy against a target URL n times.""" results = {"success": 0, "blocked": 0, "timeout": 0, "latencies": []} proxies = {"http": proxy_url, "https": proxy_url} for i in range(n): start = time.time() try: r = requests.get(target, proxies=proxies, timeout=30) latency = time.time() - start # Check for block/CAPTCHA patterns if r.status_code == 200 and "captcha" not in r.text.lower(): results["success"] += 1 results["latencies"].append(latency) else: results["blocked"] += 1 except requests.Timeout: results["timeout"] += 1 except Exception: results["timeout"] += 1 results["success_rate"] = results["success"] / n * 100 results["avg_latency"] = statistics.mean(results["latencies"]) if results["latencies"] else 0 return resultsBenchmark Results
Results averaged across all 5 target sites and all 4 geographic regions:
| Provider | Success Rate | IP Pool | Price/GB |
|---|---|---|---|
| ZentisLabs | 94.2% | 100+ MIO | from $2.39 |
| Provider A | 87.1% | 8M | $8.00 |
| Provider B | 83.7% | 5M | $9.50 |
| Provider C | 79.2% | 3M | $12.00 |
Key finding: price does NOT correlate with performance. The most expensive provider (Provider C at $12/GB) had the lowest success rate. ZentisLabs delivered the best results at less than a quarter of the price.
Per-Site Breakdown
Performance varied significantly by target site. Here's where the gaps widen:
| Site Type | ZentisLabs | Provider A | Provider B | Provider C |
|---|---|---|---|---|
| E-commerce (Akamai) | 96.1% | 89.3% | 82.1% | 76.4% |
| e-commerce Retailer | 91.8% | 81.2% | 77.4% | 70.1% |
| Travel (Cloudflare) | 95.3% | 88.7% | 86.2% | 82.3% |
| Social Media | 93.4% | 85.9% | 83.1% | 79.8% |
| Financial Data | 94.4% | 90.4% | 84.7% | 87.4% |
Why ZentisLabs Won
- Pool size matters: 100+ MIO IPs means you hit fewer previously-flagged addresses. Smaller pools get recycled faster and accumulate abuse history.
- Precise targeting: Country targeting in 195+ countries, down to state and city on Residential Light & Basic, so every request comes from exactly where the target expects its visitors.
- Regional gateways: Generated proxies connect through US, EU, or Asia gateways matched to the country you target, keeping the route short.
- Uptime: Our proxy gateways carry a 99.9% monthly uptime SLA, backed by service credits in our Terms.
How to Test Your Own Setup
Before committing to any provider, run your own benchmark:
import requests, time, statistics
def benchmark_proxy(proxy_url, n=100): """Quick proxy benchmark run before buying in bulk.""" proxies = {"http": proxy_url, "https": proxy_url} targets = [ "https://httpbin.org/ip", # Basic connectivity "https://www.google.com", # Light protection "https://www.amazon.com/robots.txt", # Medium protection ] for target in targets: latencies, successes = [], 0 print(f"\nTesting {target}...") for _ in range(n): try: t = time.time() r = requests.get(target, proxies=proxies, timeout=15) if r.status_code == 200: latencies.append(time.time() - t) successes += 1 except: pass print(f" Success rate: {successes/n*100:.1f}%") if latencies: print(f" Avg latency: {statistics.mean(latencies)*1000:.0f}ms") print(f" P95 latency: {sorted(latencies)[int(len(latencies)*0.95)]*1000:.0f}ms")
# Test ZentisLabsbenchmark_proxy("http://USER:PASS@basic.resi.zentislabs.com:12321")Always test with your actual target URLs, not just httpbin. Success rate against httpbin.org means nothing for a site protected by Akamai or Cloudflare.