Two separate #1 claims and a 500K figure, all above the foldCaptured Aug 19, 2026 · bartproxies.com
About Bart ISP
Bart Proxies is one of the louder voices in the ISP proxy space. The homepage calls them the "#1 Proxy Provider for Businesses & Individuals" and "#1 in IP Quality" (Of course, every provider claims this). They sell static residential proxies sourced from AT&T, Comcast, and Frontier, advertise a $5 trial, and claim a pool of 500K+ static IPs.
If you're deciding whether to buy, here are our initial thoughts.
The claims I can't verify
Three things on their homepage I have no way to check: the 500K+ IP count, the 115,000 customers figure, and any "#1" ranking. My benchmark covered 25 endpoints. That's enough to tell you whether the IPs are clean, not enough to count a pool or validate a customer base. I'm not implying those numbers are invented. I just can't confirm them, so I don't cite them.
One thing worth knowing, though not necessarily a contradiction: the homepage leads with "115K Satisfied Customers" and describes Bart as "trusted by 115,000 customers worldwide." Lower on the same page, their statistics section says "12,000+ Paid Monthly Users" with the line "Join a thriving community of 12,000 monthly users who already rely on us daily." Both numbers can be true at the same time: one likely counts all-time customers, the other counts people currently paying. The page never says which is which, and the number you see first is roughly ten times the one further down.
Scroll far enough and the same page says 12,000 paid monthly usersCaptured Aug 19, 2026 · bartproxies.com
What the testing actually found
Bart ranks #1 of the ISP proxies in this index, though the margin is too thin to read much into. The exact counts live in the facts panel below and update with every published run.
The IP quality claim is the one big marketing promise I can test directly, and it holds. Every exit was genuine ISP address space, usage type matched what they advertise, and zero exits came back flagged as high-risk (100/100 on trust and sourcing). If clean native carrier IPs are what you need, that part of the pitch is accurate.
Speed is good but uneven. HTTP request latency p50: 21ms. Browser page loads, on the same proxies: p50 168ms, p95 1,678ms. Throughput: 127.5 Mbps. (Their "100 GBPS Port Speed" is a port specification, not a throughput promise. My 127.5 Mbps doesn't contradict that claim.)
Rate-limit handling is the one consistent gap. On my rate-limit fixture, Bart scored 95.1%, the lowest number on their card. The other ISP proxy in this index doesn't trip that fixture at all.
The pricing situation
Their pricing page opens with the literal heading "Transparent pricing." Every price currently shown is a promo rate (50% off, code BART2026, running through August). No pre-discount price appears anywhere on the page. No billing period is named on any plan card.
The heading says transparent. Every number under it is a promo rate.Captured Aug 19, 2026 · bartproxies.com/plans/native
That's why this review doesn't print a price for them.
AvailabilityWas the proxy available when we checked?
Was the proxy available when we checked?
100
100% · 25,641 of 25,641Available intervals
NoneLongest confirmed incident
Available intervals100% · 25,641 of 25,641
How availability is graded
We check every active endpoint every five minutes. A check counts as available when the proxy authenticates, routes the request, returns our controlled response, and exposes a usable exit IP within 10 seconds. We retry once before counting a failure against the provider. Checks affected by our own controls or test system are excluded.
Product extractionCounts as complete when every one of these holds: status 200, title contains Fixture Storefront, product card and price selectors present, the text In stock present, body text at least 250 characters.
One product detail page with a breadcrumb, a product card, a price, and a stock status. It stands in for reading a single item's price and availability.
100%75 of 75 jobs completed
Search results extractionCounts as complete when every one of these holds: status 200, title contains Fixture Search Results, the result list container present, the first result's text present, body text at least 500 characters.
A search page with a query box and eight repeated result cards, each linking on to another fixture. It stands in for collecting a ranked result list and following it onward.
100%75 of 75 jobs completed
JSON API extractionCounts as complete when every one of these holds: status 200, the payload's fixture identifier present, body at least 100 characters, no selector checks because there is no DOM.
A JSON endpoint with a fixed schema and six item records, registered as a plain HTTPS target. It stands in for API scraping, and it is the one fixture with no markup to validate against.
100%75 of 75 jobs completed
Large page extractionCounts as complete when every one of these holds: status 200, title contains Fixture Large HTML, the body container present, body text at least 60,000 characters.
One page built from 160 repeated text sections, registered as a download target rather than a page target. It stands in for work whose cost is bytes rather than structure, so it is where transfer behavior shows up.
100%75 of 75 jobs completed
Listing extractionCounts as complete when every one of these holds: status 200, title contains Fixture Listing Page, grid and card selectors present, the page heading Fixture Marketplace Listings present, body text at least 900 characters.
A grid of twelve listing cards with identical shape. It stands in for pulling the same fields out of many repeated rows.
100%75 of 75 jobs completed
Completion is the share of benchmark jobs that returned the expected page data.
375 eligible observations, 0 excluded.
SpeedHow long did a valid response or completed job take?
How long did a valid response or completed job take?
98.6
How we grade this
median and p95 time to a valid job result share one log curve, with full credit at 1 second and zero at 10 seconds.
Time to a valid result on the grading curve · log scale, 100 ms to 10 s
1 s1.1 s1 s = 10010 s = 0
valid jobs p50 and p95
1 s · 375 observationsp50 time to a valid result
1.1 s · 375 observationsp95 time across eligible jobs
IP qualityWhat exit classification and reputation signals did we observe?
What exit classification and reputation signals did we observe?
100
How we grade this
mean of observed product-type match and clean-reputation rates, scaled to 0–100.
Session controlWhat identity stability and session-hold behavior did we observe?
What identity stability and session-hold behavior did we observe?
100
How we grade this
observed identity stability across comparable checks, scaled to 0–100.
100% · 50 of 50Observed identity stability
0 of 50Unexpected identity changes
Observed identity stability100% · 50 of 50
50 eligible observations, 0 excluded.
How Bart ISP performed
The grades, the baseline gate, and the price, all from our own checks, measured the same way for every proxy here.
Grade
What it measures
Scale 0–100
Bart ISP
Availability
Was the proxy available when we checked?
Bart ISP
100
Workload success
Did it complete all five scraping jobs within the retry budget?
Bart ISP
100
Speed
How long did a valid response or completed job take?
Bart ISP
98.6
IP quality
What exit classification and reputation signals did we observe?
Bart ISP
100
Session control
What identity stability and session-hold behavior did we observe?
Bart ISP
100
Baseline checks
The plumbing every proxy has to get right before anything is graded: it connects, it reports an exit IP, it serves HTTP, and its ASN, country, and download all check out. Failing one keeps a proxy off the board; passing them all never raises a grade.
These are the commodity checks nearly every serious provider passes (each needs at least 99% success over 30 or more attempts). They gate whether a proxy gets graded at all and never inflate the grades above.
passed 6 of 6
Five benchmark dimensions
Scores run from 0 to 100. Each result shows its coverage and calculation.
AvailabilityWas the proxy available when we checked?
Was the proxy available when we checked?
100
100% · 25,641 of 25,641Available intervals
NoneLongest confirmed incident
Available intervals100% · 25,641 of 25,641
How availability is graded
We check every active endpoint every five minutes. A check counts as available when the proxy authenticates, routes the request, returns our controlled response, and exposes a usable exit IP within 10 seconds. We retry once before counting a failure against the provider. Checks affected by our own controls or test system are excluded.
Product extractionCounts as complete when every one of these holds: status 200, title contains Fixture Storefront, product card and price selectors present, the text In stock present, body text at least 250 characters.
One product detail page with a breadcrumb, a product card, a price, and a stock status. It stands in for reading a single item's price and availability.
100%75 of 75 jobs completed
Search results extractionCounts as complete when every one of these holds: status 200, title contains Fixture Search Results, the result list container present, the first result's text present, body text at least 500 characters.
A search page with a query box and eight repeated result cards, each linking on to another fixture. It stands in for collecting a ranked result list and following it onward.
100%75 of 75 jobs completed
JSON API extractionCounts as complete when every one of these holds: status 200, the payload's fixture identifier present, body at least 100 characters, no selector checks because there is no DOM.
A JSON endpoint with a fixed schema and six item records, registered as a plain HTTPS target. It stands in for API scraping, and it is the one fixture with no markup to validate against.
100%75 of 75 jobs completed
Large page extractionCounts as complete when every one of these holds: status 200, title contains Fixture Large HTML, the body container present, body text at least 60,000 characters.
One page built from 160 repeated text sections, registered as a download target rather than a page target. It stands in for work whose cost is bytes rather than structure, so it is where transfer behavior shows up.
100%75 of 75 jobs completed
Listing extractionCounts as complete when every one of these holds: status 200, title contains Fixture Listing Page, grid and card selectors present, the page heading Fixture Marketplace Listings present, body text at least 900 characters.
A grid of twelve listing cards with identical shape. It stands in for pulling the same fields out of many repeated rows.
100%75 of 75 jobs completed
Completion is the share of benchmark jobs that returned the expected page data.
375 eligible observations, 0 excluded.
SpeedHow long did a valid response or completed job take?
How long did a valid response or completed job take?
98.6
How we grade this
median and p95 time to a valid job result share one log curve, with full credit at 1 second and zero at 10 seconds.
Time to a valid result on the grading curve · log scale, 100 ms to 10 s
1 s1.1 s1 s = 10010 s = 0
valid jobs p50 and p95
1 s · 375 observationsp50 time to a valid result
1.1 s · 375 observationsp95 time across eligible jobs
IP qualityWhat exit classification and reputation signals did we observe?
What exit classification and reputation signals did we observe?
100
How we grade this
mean of observed product-type match and clean-reputation rates, scaled to 0–100.
Session controlWhat identity stability and session-hold behavior did we observe?
What identity stability and session-hold behavior did we observe?
100
How we grade this
observed identity stability across comparable checks, scaled to 0–100.
100% · 50 of 50Observed identity stability
0 of 50Unexpected identity changes
Observed identity stability100% · 50 of 50
50 eligible observations, 0 excluded.
How we ran this test
One rolling 30-day window; every number on this page comes from these runs.
25endpoints tested
3eligible runsA run counts as eligible when it finished and returned at least one check the proxy itself answered. Runs that never completed, and runs where only our own test harness failed, are left out.
975checks completed
2vantage regions · graded from us-eastWhere our test machines dialled from. Grades, speed included, are measured from the graded region only; other regions are reported beside the grades, never averaged into them.
Attempt accounting
Checks passed
975 of 975
Checks that failed inside our own test harness are excluded before grading.
The full test record for this window is published as a dated report.
