We test every twenty minutes.
Here's what we find.
A Texas test client connects through GlassBox VPN, Mullvad, and Proton in turn, hits ten real websites, and logs the timing data. The script doesn't pick winners. It runs around the clock. The data on this page is live, the methodology is published, and the test harness is on Forgejo so you can run it yourself.
The last 24 hours of TTFB
Time-to-first-byte is the metric you feel. Lower is faster. The dashed line is the no-tunnel baseline (the floor any VPN is measured against).
Every twenty minutes, the Texas test client connects through each of those VPN servers in turn and measures TCP connect time, TLS handshake time, time-to-first-byte, HTTP status, and small-asset transfer against all ten websites. Plus tunnel-level metrics like round-trip time, jitter, packet loss, MTU, IPv6 leak, and DNS resolution.
We don't pick which Proton server. The harness rotates through three Proton servers in each region, cycle after cycle. Same for Mullvad. Over a single day, every Proton tunnel gets measured in every region, against every site. We're not comparing GlassBox VPN to Proton's worst server. We're comparing GlassBox VPN to all of Proton, region for region.
This is a live chart, not a screenshot. New data lands every twenty minutes.
The headline number: --% faster than Proton
Median across pooled samples, all regions, last 24 hours. Numbers update live.
These are live numbers, not a static report. They change as new test data comes in.
You feel TTFB on every page load, every API call, every script your browser fetches. It's the metric your gut registers as "the VPN is slow today." We picked it because we cannot lie about it: the Texas test client measures it from outside the tunnel, the numbers above come from the same JSON file you can curl yourself, and the test runs whether the results are good or bad for us.
Across 24 hours of testing against real, recognizable websites (Amazon, Apple, Chase, Microsoft, Yahoo, GitHub, Wikipedia, Stack Overflow, BBC, Craigslist), GlassBox VPN's median TTFB sits well ahead of Proton's in every region we test. The gap holds whether the test ran at 3 AM or 3 PM. It holds across all ten sites. It holds whether we run the next cycle in five minutes or tomorrow.
Real websites. Real differences.
Median TTFB per site, last 24 hours, all regions pooled. Five highlighted below. The full ten-site breakdown is in the raw JSON.
This is a live chart, not a screenshot. New data lands every twenty minutes.
Site selection is fixed and pre-registered in the methodology. We don't pick favorable sites. The test runs against the same ten every twenty minutes regardless of how the numbers come out. Full per-region per-site data is in the raw JSON, including Mullvad's full breakdown alongside GlassBox VPN's and Proton's.
About Mullvad
Mullvad ties us. Not approximately. Statistically.
Across all regions, all ten sites, every twenty-minute cycle, day in and day out, Mullvad's medians and GlassBox VPN's medians sit on top of each other. Sometimes we're ahead by a millisecond or two. Sometimes they are. The gap is noise. That is not luck. That is two different teams building the same thing the right way and arriving at the same answer.
Mullvad runs disciplined infrastructure and has for years. They earned every recommendation they get and we will not pretend otherwise just because we compete with them.
Full Mullvad data is in the raw JSON alongside GlassBox VPN's and Proton's.
Why this matters past the speed test
The next decade of VPNs belongs to small operators that tell the truth.
Two or three giant VPN brands captured the consumer market and used that scale to make claims that increasingly read like legal dares. Maximize privacy. Bypass everything. Promise the moon. It worked, until governments noticed, and now we're watching laws pass in Europe and elsewhere that wouldn't have been written if the category had stayed small and quiet and honest.
The original promise of a VPN was simpler: surf the internet without your ISP and ad networks selling your traffic. That promise doesn't require a fortress. It requires a working tunnel, a clear policy on what is logged, and an operator small enough to stay out of the regulatory bullseye. Decentralized. Many small operators. None of them big enough to be the next bullet point in a piece of legislation.
GlassBox VPN is built on that bet. Small. Transparent. US-only. We're not chasing the giants, we're betting against them. If you want a VPN industry that survives the next decade in a form worth using, the move is to support the small operators that do it right. Mullvad is one. GlassBox VPN is one. There are others. Find them. Subscribe to the ones that earn it. That's how the industry stays useful.
GlassBox VPN donates 5% of subscription revenue to the WireGuard project. The protocol is what makes any of this fast. We pay the people who built it.
Methodology and replication
Don't trust our charts. Run the test yourself.
The Texas test client runs from a server with no ties to any VPN provider. It connects through one of three GlassBox VPN servers, one of nine Proton servers, or one of nine Mullvad servers, then makes HTTP requests to the same ten websites and logs the timing data. Cycles run every twenty minutes, around the clock. The methodology specifies every choice in advance: site list, percentile aggregation, how we handle errors, the conflict-of-interest disclosure.
Everything is on Forgejo. Test harness, the Docker compose that runs it, the SQLite schema, the snapshot emitter that produces the JSON this page reads. If you want to run the test yourself with your own VPN providers and your own client location, the code is yours.
Test Harness Source
Docker compose, test runner, snapshot emitter. Open source. Replicate it on your own hardware.
Forgejo →Raw JSON
The exact data this page renders. No transforms, no editorial. Curl it, parse it, check our math.
/proof/data.json$60 a year. WireGuard. No logs. Verified.
GlassBox VPN is a single-peer WireGuard VPN for one device. US-only servers, 21+ adults only, blocklist on every server, configs on Forgejo, and live test data you can drill into right now.