Benchmarks

Numbers we can defend.

This page separates internal observations from published evidence. The raw harness, hardware list, and run data are still being prepared; until they are public, treat the figure below as an internal measurement, not an independently reproducible benchmark.

Current evidence

One internal result, one documented limit.

internal bCloud multipart upload → 3.6× (1.8 → 6.4 MB/s)free relay → 1 MiB/s, by design

bCloud multipart upload: 3.6× in an internal run. The bCloud HTTP upload path moved data at 1.8 MB/s with one stream and 6.4 MB/s with four concurrent multipart streams. This does not measure direct peer-to-peer QUIC/BBR performance. The raw reproduction package is still pending.
Free relay: 1 MiB/s.Not a measurement — a cap. The free plan’s relay fallback is throttled to 1 MiB/s on purpose; direct transfers are not throttled on any plan.

Publication status

What is documented, and what is still pending.

Scope. The 3.6× figure compares one stream with four concurrent streams on the bCloud HTTP multipart upload path.
Boundary. It is not a direct-transfer comparison and should not be used as a QUIC-vs-TCP or Blixt-vs-competitor result.
Reproduction package in progress. The full harness, hardware list, network conditions, repetition count and raw runs will be published before we call the result independently reproducible.

Illustrative transfer-time races have been removed from the landing page. Independent reproduction is welcome once the inputs and raw runs are public.

In progress: a full cross-service shootout — Blixt vs scpvs the cloud services, on three scenarios (1 Gbit/s LAN, city↔city, and an intercontinental link with real loss). It will be published here, including any results that don’t flatter us. No ETA promises — it ships when the method holds up.

Want to try it on your own link in the meantime? Download Blixt — the built-in speed test takes a minute.