Benchmark

What we measured, how we measured it, and where we lose.

Xebit holds 9,200 concurrent one to one voice sessions per node against 2,800 for LiveKit (open source) on identical hardware, a 3.3x difference. P95 mouth to ear latency is 118 ms against 142 ms. On participants in a single room we are lower, 82 against 780. Every figure below comes with the method that produced it, and we will set you up to reproduce all of it on your own machines.

Updated

How does Xebit compare with a reference voice framework?

Nine measurements on identical hardware. Lower is better for time, higher is better for capacity and MOS.
Xebit against LiveKit (open source), identical hardware
MetricXebitLiveKitDifference
Concurrent one to one sessions per node9,2002,8003.3x
Participants per node, all one to one18,4005,6003.3x
Agent Server sessions, dedicated node10,50018see note
Median join to audio, cold join191 ms247 ms−23%
P95 mouth to ear, two party call118 ms142 ms−17%
SIP call setup, PSTN ingress430 ms580 ms−26%
Reconnect after forced node kill1.3 s2.2 s−41%
Audio MOS at 8% packet loss3.93.6higher is better
Participants in a single room82780we are lower
The Agent Server row is measured on its own dedicated node because it has a different resource profile, and it is not additive with the Bridge rows. The LiveKit figure is the last stable step before that node exhausted memory.

Why is Xebit worse at large rooms?

On purpose.

A conferencing SFU is tuned for large rooms with a few active speakers. A voice agent platform is thousands of isolated two party sessions where someone is talking in every single one. Those are opposite optimisation targets in the media path and in per session state, and you cannot have both.

We chose the second, which is why we lead the one to one rows by roughly three times and lose the single room row by roughly nine. If your use case is a 780 person town hall, use something else. If it is voice agents at volume, this trade is the entire point.

How was each number produced?

Latency

RTP timestamp echo. A monotonic clock embedded in the audio payload, so the figure is a true end to end audio path measurement rather than a signalling round trip. Median and P95 across 500 samples per run.

Packet loss

Injected with tc and netem at 2 to 10 percent random loss on the sender path. Scored with a PESQ derived MOS on the recorded output, so the score comes from the received audio and not from a loss counter.

SIP

Measured from SIP INVITE at a real carrier test trunk to the first inbound RTP packet observed at the Bridge, averaged over 100 calls. A real trunk, not a loopback.

Failover

An active call is sustained while the serving SFU process is forcibly terminated. Reconnect is measured from the disconnect event to audio resumption at the client SDK.

Concurrency

The harness ramps concurrent isolated two party calls rather than growing a single room. Participant count is reported as twice the maximum stable session count.

Repetition and instrumentation

Every test repeated five times and reported as a trimmed mean, with the top and bottom runs excluded. Metrics collected with Beats on each node and shipped to an ELK stack, so instrumentation lives outside the system under test.

Does it hold in production?

A benchmark is a lab. This is a month of real traffic.
2.5Mcalls in 20 days
150Kcalls a day, sustained all month
67%pickup and engagement
1VM

In April 2026 we ran a national political campaign in India: 2.5 million calls in 20 days, 150,000 a day sustained for the full month, at 67% pickup and engagement, on a single VM. The compute bill for that month was about $800. It was a one time project.

Can I reproduce these results?

Yes. Tell us the concurrency, cost and response time you need and we will set up an environment on your hardware so you can check every figure on this page. The harness, the configuration and the measurement tooling come with it.