Skip to main content
vpntrackr.
2026 Report

The VPN Speed Report.

6,200 hours of testing across 28 providers, 5 cities, and 3 protocols. The data behind our 2026 ranking — fully open.

Published 2026-05-16VpnTrackr Editorial12 min read
Providers tested
28
Hours of testing
6,200h
Cities in the panel
5
Protocols compared
3
Methodology

How the report was built.

Speed reports are notoriously gameable. This one was structured to be reproducible — same rig, same windows, same baseline.

  1. Step 01

    Same rig, every test.

    Two identical desktops — Core i7-14700, 32GB DDR5, NVMe storage — running Windows 11 and Ubuntu 24.04 in parallel on a 1 Gbps symmetric fibre baseline. The baseline is re-verified before every run; tests pause if it drifts more than 3% from the rolling weekly median.

  2. Step 02

    Three protocols, every cycle.

    Every provider is run on WireGuard (or its derivative), OpenVPN UDP, and IKEv2/IPsec where supported. Same MTU, same DNS, same kill-switch state. Speed is the median of the protocol matrix, not the peak — that discourages cherry-picking.

  3. Step 03

    Five-city panel.

    Frankfurt, New York, Singapore, London, Sydney. Speed downloads are uncompressible random bytes served from our own endpoint to defeat any traffic shaping or transparent cache that might inflate a marketing-friendly number.

  4. Step 04

    Reproducible by anyone.

    Every protocol setting, every server-pick rule, every tolerance is published. A provider who believes a score is wrong can reproduce the test using the published rig spec. When their numbers differ, we re-run and publish both dated.

Want the full document?

The methodology is also published in long form alongside the rest of our scoring rubric, including the privacy, streaming, and value pillars and how they roll up into an overall score.

Read the full methodology →
Findings

The headline numbers.

Where providers separated from the pack, which protocols won where, and what the variance looked like over the 14-day panel.

≈92%
Top-quartile speed retention

The best-performing providers retain roughly 92% of the unencrypted baseline on regional routes. The bottom quartile loses more than half.

WireGuard
Won every region

WireGuard (and the NordLynx / Lightway derivatives) was the fastest protocol in every panel city, every test cycle. OpenVPN UDP is now strictly a fallback.

13×
Variance, slowest to fastest

The slowest provider in the panel returned roughly 1/13th the throughput of the fastest on long-haul routes. Brand reputation was a poor predictor of where a provider landed in that spread.

Reading the panel

The headline ranking on every provider review is built from the same protocol matrix described above. The per-VPN page surfaces the median throughput and percentage of the unencrypted baseline retained — those are the two numbers that move the speed sub-score.

See the ranking →
Raw data

The dataset.

Every measurement, every protocol, every region. Cleaned for release alongside this report.

The full per-provider CSV, the analysis notebook, and the protocol-by-protocol JSON dump are being prepared for public release. We’re doing one final pass for personally-identifying fields (none expected — the dataset is throughput numbers, not users) and adding a permissive licence so researchers and journalists can cite freely.

In the interim, the per-provider numbers that feed the public ranking are live on every review page. If you need the raw panel for academic or editorial work before the public release, write to the editorial desk and we’ll share the current snapshot.

See per-provider resultsRequest the raw dataset
Providers in this report

Browse the rankings.