Proxy Benchmark 2026: Seven Residential Providers Tested on AI Agents

Seven rotating residential proxy providers were benchmarked on browser-agent tasks on November 3, 2026, and entry prices now start as low as $3.50 for 1 GB without a subscription. Decodo and DataImpulse came out cleanest on the hardest test: both restored and submitted a saved draft after a five-minute idle pause in two out of two controlled sessions, while Oxylabs, FloppyData, Byteful and Webshare each logged one failure in two.

The test, run with Browser Use's hosted browser, bu-fast, US residential exits and fresh sessions, covered product research on Amazon and eBay, full-text extraction, form filling, catalog-to-form workflows, and session continuity. Every provider scored 3/3 on Amazon. eBay was the weak spot: only Decodo, Oxylabs and FloppyData managed 1/3, and a separate eBay continuity check failed initial access for every configuration and was excluded. Roundproxies, Byteful, DataImpulse and Webshare all scored 0/3 on eBay.

Full-text extraction split the field. Decodo, FloppyData, DataImpulse and Webshare hit 3/3; Oxylabs and Byteful managed 2/3; Roundproxies got 1/3. The testers attribute those failures to model refusals after page access, not blocked connections. All seven providers completed forms at 3/3 and catalog-to-form runs at 2/2, scored against server receipts.

What does the pricing look like?

Entry prices, checked November 5, 2026, vary widely in structure even when the headline rates look similar:

• Decodo — 115M-IP pool (provider claim), targeting down to country, city, ZIP and network, sticky sessions up to 24 hours. $4 for 1 GB with no subscription, or from $11.25/month for 3 GB. A 100 MB trial lasts three days and requires a payment method. Traffic expires after 12 months.
• Oxylabs — 175M IPs, unlimited concurrent sessions advertised. Starter costs $15/month in shared credits with residential traffic at $6/GB; unused credits expire each cycle, and traffic is billed even for destination errors.
• FloppyData — 65M IPs, one traffic balance across residential, mobile and datacenter proxies. $19 for 19 GB (30 days) or $28.50 non-expiring. Documented ten-minute idle expiry; in one of two checks the IP changed after five idle minutes, though the saved draft was restored.
• Roundproxies — 119M IPs, country and city targeting. $3.50 for 1 GB, prepaid, no expiry. A 500 MB, 24-hour trial is advertised but availability needs confirming.
• Byteful — 35M IPs, country, region, city and ISP targeting. $3.25/month for 1 GB; bandwidth never expires, even after cancellation, but the plan renews monthly. The advertised 1 GB trial's eligibility terms are contradicted by Byteful's own pricing FAQ.
• DataImpulse — 90M IPs, country targeting included, sticky sessions up to two hours. $5 for 5 GB with no expiry, but later top-ups of the same proxy type require at least $50.
• Webshare — 80M IPs, HTTP and SOCKS5, targeting through country, state, city, ZIP and network. $3.50/month for 1 GB, no one-off option. The free ten-proxy plan uses datacenter IPs and won't test residential access.

Global IP totals are unverified provider claims, the testers note, and features follow provider documentation.

Should you buy proxies separately or with the browser?

Browser platforms sell bundled residential proxies at rates from $5/GB to roughly $11/GB. Browser Use charges $5/GB pay-as-you-go with no subscription; Browserbase runs about $11/GB (Developer: $20/month with 1 GB included, then $12/GB); Browserless about $10.50/GB; Steel about $8/GB (Scale plan drops to $6/GB at $250/month); Hyperbrowser $10/GB; Kernel includes proxy cost in browser usage. Prices were checked November 6, 2026.

Bring-your-own matters most at scale — with some providers, proxy traffic costs more than the browser itself, though small workflows may not notice.

What should buyers check before committing?

The testers list four checks: session continuity (sticky sessions don't guarantee a reserved IP, and cookies plus storage matter as much as the address), location coverage and concurrency limits, access to the actual target sites (browser fingerprinting exposes signals beyond the IP, so switching providers won't fix every failure), and cost per completed task — retries, failed pages and expired traffic can erase a cheap per-GB rate. The sample sizes here are small and differently timed, and the testers say the results don't establish a universal reliability or speed ranking.

The practical next step the testers recommend: run a few providers on your real workflow with pauses and recovery included, and compare cost per successful task rather than price per gigabyte.